append completed text2026070617374416c893e52f89

data.jsonl

@@ -2941,3 +2941,4 @@
 {"id": "ner202603121539318bc69e3f196c", "timestamp": "2026-03-12T15:39:33.509433+00:00", "service": "ner", "event_type": "completed", "status": "completed", "progress": 100, "current_step": null, "filename": "text_input_txt20260312153931a9f79e3602fe.txt", "input_type": "text", "start_page": null, "end_page": null, "total_pages": null, "steps": ["ner"], "input_file_path": null, "output_file_path": "https://point9ml.s3.amazonaws.com/outputs/output_ner202603121539318bc69e3f196c.json?response-content-type=application%2Fjson&response-content-disposition=attachment%3B%20filename%3D%22output_ner202603121539318bc69e3f196c.json%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260312%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260312T153934Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=f89bd01b6088b83896cc05a0fed60555474aec3f292d5a6259ea23c5d790064e", "created_at": "2026-03-12T15:39:31.730924+00:00", "updated_at": "2026-03-12T15:39:33.509433+00:00", "finished_at": "2026-03-12T15:39:33.509433+00:00", "duration_ms": 1778, "error_message": null, "output": {"process_id": "ner202603121539318bc69e3f196c", "status": "completed", "filename": "text_input_txt20260312153931a9f79e3602fe.txt", "result": "[NER Error Chunk 1] {\"error\":{\"message\":\"This request requires more credits, or fewer max_tokens. You requested up to 2048 tokens, but can only afford 2047. To increase, visit https://openrouter.ai/settings/credits and upgrade to a paid account\",\"code\":402,\"metadata\":{\"provider_name\":null}},\"user_id\":\"user_2ylQWLYghssJxdVaY8yayMePOMx\"}", "input_file_path": null, "output_file_path": "https://point9ml.s3.amazonaws.com/outputs/output_ner202603121539318bc69e3f196c.json?response-content-type=application%2Fjson&response-content-disposition=attachment%3B%20filename%3D%22output_ner202603121539318bc69e3f196c.json%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260312%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260312T153934Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=f89bd01b6088b83896cc05a0fed60555474aec3f292d5a6259ea23c5d790064e"}}
 {"id": "text2026070617374416c893e52f89", "timestamp": "2026-07-06T17:37:44.667509+00:00", "service": "extract_text", "event_type": "created", "status": "queued", "progress": 0, "current_step": null, "filename": "Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_type": "file", "start_page": 1, "end_page": 16, "total_pages": 16, "steps": ["extract_text"], "s3_key": "documents/Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_file_path": "https://point9ml.s3.amazonaws.com/documents/Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf?response-content-type=application%2Fpdf&response-content-disposition=attachment%3B%20filename%3D%22Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173749Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=683cccfdb1f09ccadfafa674c9f05ffb74fcbab71e4458da55d88bbf640bc284", "output_file_path": null, "created_at": "2026-07-06T17:37:44.667509+00:00", "updated_at": "2026-07-06T17:37:44.667509+00:00", "finished_at": null, "duration_ms": null, "error_message": null}
 {"id": "text2026070617374416c893e52f89", "timestamp": "2026-07-06T17:37:55.368291+00:00", "service": "extract_text", "event_type": "progress", "status": "in_progress", "progress": 99, "current_step": "extract_text", "filename": "Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_type": "file", "start_page": 1, "end_page": 16, "total_pages": 16, "steps": ["extract_text"], "s3_key": "documents/Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_file_path": "https://point9ml.s3.amazonaws.com/documents/Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf?response-content-type=application%2Fpdf&response-content-disposition=attachment%3B%20filename%3D%22Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173749Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=683cccfdb1f09ccadfafa674c9f05ffb74fcbab71e4458da55d88bbf640bc284", "output_file_path": "https://point9ml.s3.amazonaws.com/outputs/text_text2026070617374416c893e52f89.txt?response-content-type=text%2Fplain&response-content-disposition=attachment%3B%20filename%3D%22text_text2026070617374416c893e52f89.txt%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173755Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=a61f9698f9422da9c515a2f2f7e97b31b23f591567923cd0e9c62743db80e642", "updated_at": "2026-07-06T17:37:55.368325+00:00"}
+{"id": "text2026070617374416c893e52f89", "timestamp": "2026-07-06T17:37:57.639895+00:00", "service": "extract_text", "event_type": "completed", "status": "completed", "progress": 100, "current_step": null, "filename": "Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_type": "file", "start_page": 1, "end_page": 16, "total_pages": 16, "steps": ["extract_text"], "s3_key": "documents/Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "input_file_path": "https://point9ml.s3.amazonaws.com/documents/Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf?response-content-type=application%2Fpdf&response-content-disposition=attachment%3B%20filename%3D%22Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173749Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=683cccfdb1f09ccadfafa674c9f05ffb74fcbab71e4458da55d88bbf640bc284", "output_file_path": "https://point9ml.s3.amazonaws.com/outputs/text_text2026070617374416c893e52f89.txt?response-content-type=text%2Fplain&response-content-disposition=attachment%3B%20filename%3D%22text_text2026070617374416c893e52f89.txt%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173755Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=a61f9698f9422da9c515a2f2f7e97b31b23f591567923cd0e9c62743db80e642", "created_at": "2026-07-06T17:37:44.667509+00:00", "updated_at": "2026-07-06T17:37:57.639895+00:00", "finished_at": "2026-07-06T17:37:57.639895+00:00", "duration_ms": 12972, "error_message": null, "output": {"process_id": "text2026070617374416c893e52f89", "status": "completed", "filename": "Machine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision treatment _ PLOS Computational Biology.pdf", "result": "--- Page 1 ---\nAbstract\nThe heterogeneity of Hepatocellular Carcinoma (HCC) poses a barrier to effective treatment. Stratifying highly heterogeneous HCC\ninto molecular subtypes with similar features is crucial for personalized anti-tumor therapies. Although driver genes play pivotal\nroles in cancer progression, their potential in HCC subtyping has been largely overlooked. This study aims to utilize driver genes to\nconstruct HCC subtype models and unravel their molecular mechanisms. Utilizing a novel computational framework, we expanded\nthe initially identified 96 driver genes to 1192 based on mutational aspects and an additional 233 considering driver dysregulation.\nThese genes were subsequently employed as stratification markers for further analyses. A novel multi-omics subtype classification\nalgorithm was developed, leveraging mutation and expression data of the identified stratification genes. This algorithm successfully\ncategorized HCC into two distinct subtypes, CLASS A and CLASS B, demonstrating significant differences in survival outcomes.\nIntegrating multi-omics and single-cell data unveiled substantial distinctions between these subtypes regarding transcriptomics,\nmutations, copy number variations, and epigenomics. Moreover, our prognostic model exhibited excellent predictive performance in\ntraining and external validation cohorts. Finally, a 10-gene classification model for these subtypes identified TTK as a promising\ntherapeutic target with robust classification capabilities. This comprehensive study provides a novel perspective on HCC\nstratification, offering crucial insights for a deeper understanding of its pathogenesis and the development of promising treatment\nstrategies.\nAuthor summary\nDividing highly heterogeneous HCC into molecular subtypes with similar characteristics is crucial for personalized anti-tumor\ntherapies. Although driver genes play pivotal roles in cancer progression, their potential in HCC subtyping has been largely\noverlooked. In this work, we developed a multi-omics network-based stratification algorithm that utilizes patient mutation data and\nrequires smaller computational resources for subtype assignment. Through this algorithm, we categorized HCC into two subtypes,\nCLASS A and CLASS B. Using multi-omics and single-cell data, we identified differences between these subtypes in gene\nexpression, methylation, immune infiltration, and other aspects. Beyond subtype characterization, our study established a robust\nclinical prediction model (https://mike-wang-bjut.shinyapps.io/DynNomapp_HCC_Sutypes/) incorporating subtype information and\ntypical clinical features, enabling precise survival predictions. Finally, we developed a high-performing machine learning classifier\nfor our subtype. Analyzing this classification model and reviewing previous experimental papers, we identified TTK as a potential\ndiagnostic marker and therapeutic target specific to our subtypes. In conclusion, our research offers a novel perspective on HCC\nstratification, which is crucial for a deeper understanding of its pathogenesis and developing promising treatment strategies.\nCitation: Wang M, Yan X, Dong Y, Li X, Gao B (2024) Machine learning and multi-omics data reveal driver gene-based\nmolecular subtypes in hepatocellular carcinoma for precision treatment. PLoS Comput Biol 20(5): e1012113.\nhttps://doi.org/10.1371/journal.pcbi.1012113\nEditor: Stacey D. Finley, University of Southern California, UNITED STATES\nReceived: January 2, 2024; Accepted: April 24, 2024; Published: May 10, 2024\nCopyright: © 2024 Wang et al. This is an open access article distributed under the terms of the Creative Commons\nAttribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author\nand source are credited.\nData Availability: The data used in this study are all from public databases TCGA\n(https://www.cancer.gov/ccg/research/genome-sequencing/tcga) and ICGC (https://dcc.icgc.org/). The Fudan cohort and\nsingle-cell dataset are from the GEO database with accession numbers GSE14520 and GSE149614, respectively. The\nLIMORE dataset was obtained from its original paper (PMID: 31378681). To facilitate clinical translation, we developed an\ninteractive HCC prognosis model: https://mike-wang-bjut.shinyapps.io/DynNomapp_HCC_Sutypes/. The subtype classifiers\nSVM_10 and SVM_TTK, along with the relevant code, are stored at https://github.com/Mike-\nW29/SVM_model_for_HCC_subtype.\nFunding: This study was supported by the National Key Research and Development Program of China (2022YFC2704804 to\nBG), and the National Natural Science Foundation of China (61931013 to BG). The funders had no role in study design, data\ncollection and analysis, decision to publish, or preparation of the manuscript.\nCompeting interests: The authors declare that they have no competing interests.\nIntroduction\nHepatocellular carcinoma (HCC) is recognized as the most prevalent primary liver malignancy, ranking as the third leading cause of\ncancer-related deaths globally and experiencing a notable increase in incidence [1,2]. The molecular and pathological\nheterogeneity of HCC presents a formidable obstacle to developing personalized therapeutic approaches [3]. Therefore, using key\nfeatures to classify different HCC patients into relatively homogeneous subtypes is clinically essential.\nMachine learning and multi-omics data reveal driver gene-\nbased molecular subtypes in hepatocellular carcinoma for\nprecision treatment\nVersion 2\nPublished: May 10, 2024\nhttps://doi.org/10.1371/journal.pcbi.1012113\nMeng Wang, Xinyue Yan, Yanan Dong, Xiaoqin Li\n, Bin Gao\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n1/16\n\n--- Page 2 ---\nRecent advances in high-throughput sequencing technologies have facilitated the comprehensive profiling of patient dysfunctions\nacross multiple biological systems. Through various omics techniques, potential oncogenic factors can be discerned [4]. Utilizing\nbig data in molecular subtyping of HCC has become increasingly feasible [5]. For instance, Zhang et al. employed mass cytometry\ndata to categorize HCC into three subtypes, each exhibiting diverse immune activities [6]. Poirion et al. developed the DeepProg\ndeep learning method, which integrates RNA, DNA methylation, and miRNA data to classify HCC into two subtypes with distinct\nsurvival differences and biological profiles [7].\nIn tumors, driver genes are often causally related to tumor progression. Compiling a complete list of driver genes is crucial for\noncology diagnosis and drug development [8]. We have identified recurrently altered driver genes in HCC, and some of these\ngenes have been suggested to be associated with specific molecular subtypes [9]. However, no systematic subtyping studies of\nHCC using driver genes exist to our knowledge.\nGene families, representing a cluster of genes with shared ancestry and similar biochemical functions, present an opportunity to\nidentify rare carcinogenic mutations [10–12]. In this study, we utilized gene families to expand the list of driver genes, culminating in\ncreating two distinct molecular subtypes for HCC. By integrating multi-omics data and single-cell information, we explored the\nunique characteristics defining these subtypes, spanning transcriptomics, genomics, epigenomics, immune infiltration, and tumor\nstem cell activity. Additionally, we developed an interactive prognostic model website for the two subtypes, empowering users to\neffortlessly generate personalized survival predictions based on critical patient information, including age, stage, virus infection\nstatus, and subtype classification. Finally, we established a 10-gene classification model for these subtypes and singled out TTK as\na promising therapeutic target with strong classification capabilities. In summary, our study provides a fresh perspective on the\nconstruction of HCC subtypes and offers promising avenues for future therapeutic strategies.\nResult\nObtaining stratification genes from driver genes and their family members\nTo ensure a comprehensive and high-confidence selection of driver genes, we gathered HCC driver gene lists from three distinct\nstudies. Bailey et al. [13] employed diverse driver gene discovery algorithms and conducted meticulous manual curation to\nconstruct their driver gene list. Martínez-Jiménez et al. [14] extended the scope by analyzing a larger sample size and adopting a\nmore comprehensive approach to exploring driver genes. Meanwhile, Fujimoto et al. [15] concentrated on HCC, providing valuable\ninsights into the specific driver mechanisms of HCC. By amalgamating the findings of these three studies, we curated a list of 96\nHCC driver genes for our subsequent analyses. Furthermore, we enriched this selection by including their corresponding family\nmembers, which were sourced from InterPro (https://www.ebi.ac.uk/interpro/), UniProtKB (https://www.uniprot.org/), as well as\nseveral other references [16–21]. (S1 Table).\nTo identify protein domains with significant mutation burden, we first annotated the domains for each gene using the PfamScan\n(https://www.ebi.ac.uk/Tools/pfa/pfamscan/) and excluded those with an e-value greater than 1e-5. Additionally, considering the\npotential role of the Degron region in transcription factors for cancer growth [22], we included Degron as a protein domain based on\nDegpred predictions [23]. Through permutation testing, we identified 75 protein domains with significant mutation burden (Fig 1A).\nFocusing on protein domains with higher entropy values, which may indicate novel oncogenic alterations, we identified 1192\nstratification genes mutated in domains with entropy values greater than 0.5 (S2 and S3 Tables).\nFig 1. illustrates the process of obtaining stratification genes.\n(A) The workflow for protein domains with significant mutation burden (B) illustrates the process of defining DDGs.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g001\nThe progression of HCC involves genetic mutations, epigenetic variations, and dysregulated gene expression [24–26]. Many well-\nknown HCC driver gene alterations are associated with an epigenetic variation or copy number variation (CNV). To complete the\nstratification genes, we conducted differential analysis between normal and tumor tissues, identifying 244 differentially expressed\ngenes within gene families. Then, we integrated other omics data to define 223 driver-dysregulated genes (DDGs) (Fig 1B and S3\nTable).\nDriver gene-related HCC Subtypes\nWe stratified HCC into distinct subtypes using mutation data from stratification genes and DDGs expression data. Firstly, we\nsmoothed the patient mutation matrix using the Network-based stratification (NBS) algorithm [27]. Next, we integrated the\nsmoothed mutation data with DDGs expression data using the Similarity network fusion (SNF) algorithm to construct a similarity\nmatrix among samples [28]. The amalgamation of SNF with a consensus cluster facilitated patient subtype assignment, yielding\nmore robust and desirable clustering outcomes (Fig 2A, Table 1). The combination of smoothed mutation data and DDG expression\ndata proved effective in capturing information and enhancing the clustering analysis.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n2/16\n\n--- Page 3 ---\nFig 2. Identification of driver gene-related subtypes.\n(A) Flowchart depicting the process of the subtype classification algorithm; (B) Silhouette coefficients for different values of k;\n(C) Silhouette plot specifically for k = 2; (D) Heatmap of the consensus matrix defining the two subtypes; (E) Five-year\nsurvival curves for the two subtypes, with CLASS A represented in red and CLASS B represented in blue.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g002\nTable 1. Comparison of different clustering methods.\nhttps://doi.org/10.1371/journal.pcbi.1012113.t001\nChoosing the suitable gene interaction network is crucial for the NBS algorithm’s smoothing effect. We compared two commonly\nused networks, String (Requested score = medium confidence, FDR stringency = medium) and HumanNetv3 (top 10% confidence)\n[29]. We found that HumanNetv3 performed better in stratification based on silhouette coefficients and p-values (Table 1).\nAdditionally, we examined the stratification outcomes by utilizing only driver genes versus incorporating gene family members for\nHCC. The results demonstrated that the latter approach exhibited superior performance in both clustering stability and survival\ndifferences, as reflected in silhouette coefficients and p-values (Table 1).\nTo determine the optimal cluster count, we computed silhouette coefficients for various cluster numbers (k), revealing that k = 2 was\nthe most suitable parameter for delineating HCC subtypes (Fig 2B and 2C). We designated these subtypes as CLASS A and\nCLASS B (Fig 2D), where patients with CLASS B had worse prognostic outcomes (median OS time 17 months vs 20.55 months)\n(Fig 2E). Subsequent validation of our method on the TCGA_LIHC cohort further confirmed significant survival differences between\nthe two subtypes (median OS time 15.97 months vs 22.42 months) (S1 Fig). This demonstrates that integrating smoothed mutation\ndata with gene expression data effectively classifies HCC subtypes.\nOur identified HCC subtypes significantly correlated with independently studied subtypes [30–33] (chi-square test, see S4 and S5\nTables, S2 Fig). We constructed Cox regression models based on these subtypes. We assessed predictive performance using the\nC-index and Integrated Brier Score (IBS) to evaluate the association of different subtypes with clinical prognosis. The results\ndemonstrate that our constructed subtypes excel in C-index, reaching 0.611 (95% CI = 0.587–0.635), significantly higher than most\nother models (S3 Fig). Simultaneously, the IBS is 0.183, lower than other models (Hoshida: 0.195, Bidkhori: 0.205, Benfeitas:\n0.194, TCGA: 0.180). This indicates that our subtypes exhibit higher consistency and accuracy in predicting patient survival than\nothers.\nBiological properties of different subtypes\nTo explore distinct biological properties, we performed gene set variation analysis (GSVA) on KEGG, Reactome, Hallmark, and\noncogenic signature gene sets for the two subtypes. The results revealed high consistency in enrichment patterns across these\ngene sets for both subtypes (Fig 3A–3C and S6 Table). CLASS A showed higher GSVA scores in metabolic pathways such as\noxidative phosphorylation, organic acid metabolism, fatty acid metabolism, and glycolysis. In contrast, CLASS B displayed a\nheightened proliferative profile, enriched in pathways associated with mitosis, cell cycle, DNA replication, and cell cycle checkpoint.\nAdditionally, CLASS B had upregulated activity in pathways related to histone methylation, DNA methylation, and inflammatory\nsignaling (S6 Table).\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n3/16\n\n--- Page 4 ---\nFig 3. Different biological properties of the two subtypes.\n(A-C) Differential analysis of KEGG, Hallmark and oncogenic signature pathways between these two subtypes. (D) The\nimmune cell abundance in these two subtypes using TIMER, with statistical significance assessed by the Mann-Whitney U\ntest. (E) Box plots depicting the expression levels of six immune checkpoint genes in these two subtypes, with red boxes\nindicating significantly differentially expressed genes (p.adjust < 0.05, |log2FC| > 1). (F) Box plot showing the mRNA\nstemness scores (mRNAsi) of the two subtypes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g003\nNext, we analyzed immune cell composition using TIMER [34] and xCell algorithms [35], revealing a higher abundance of immune\ncells, including T cells, B cells, and macrophages, in CLASS B than CLASS A. CLASS B had a higher stromal score. In contrast,\nCLASS A had a higher microenvironmental score (Figs 3D, S4A and S4B).\nWe also examined the expression of six common immune checkpoints [36,37], finding higher expression levels in CLASS B than in\nCLASS A (Fig 3E). Specifically, PD-1 (PDCD1), CTLA-4 (CTLA4), TIM-3 (HAVCR2), and TIGIT were differentially expressed in\nCLASS B, suggesting potential responsiveness to immune checkpoint inhibitors.\nFinally, we used Malta et al.’s machine learning [38] algorithm to examine the stemness features of two subtypes. The results\nshowed that CLASS B exhibited stronger stemness features, indicating a higher potential for invasion and metastasis (Fig 3F). This\nalteration may be associated with the high expression of KRT19, a marker for biliary/hepatic progenitor cells, in CLASS B (S5 Fig)\n[39].\nMulti-omics properties of subtypes\nDespite no significant difference in mutation count and TMB values between the subtypes (Figs 4A and S6), CLASS B showed\nhigher chromosomal instability (Fig 4B). Specifically, CLASS B displayed deletions in 4p, 4q, 13q, 16p, 16q, and 17p, while CLASS\nA predominantly manifested amplifications in 5q (Fig 4C).\nFig 4. Distinct multi-omics features of the two subtypes.\n(A) Box plot showing the number of nonsynonymous mutations in each subtype. (B) Box plot displaying the CIN ratio in each\nsubtype. (C) Heatmap illustrating CNVs of the 22 autosomes in both subtypes, with red and blue indicating copy number\namplifications and deletions, respectively. (D) Oncoplot presents the top 20 significantly mutated genes in the subtypes based\non the p-values. (E) Identifying of subtype-specific methylation probes in the two groups using the R package ChAMP, with a\nthreshold of p.adjust < 0.05 and |Δβ| > 0.2. (F) Expression profiles of DNA methyltransferase family members DNMT1,\nDNMT3A, and DNMT3B in these two subtypes.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n4/16\n\n--- Page 5 ---\nhttps://doi.org/10.1371/journal.pcbi.1012113.g004\nUsing the chi-square test, we identified significantly mutated genes in these subtypes (Fig 4D). TP53 and CTNNB1, well-known\nHCC driver genes, had different distributions between subtypes. Consistent with previous research, TP53 was mainly found in\nCLASS B, associated with a poorer prognosis, while CTNNB1 was mainly associated with CLASS A, linked to a better prognosis\n[40,41]. Additionally, RB1 mutations were predominantly enriched in CLASS B, possibly contributing to its high proliferative profile\n(S6 Table).\nBy analyzing TCGA data, we found significant hypermethylation patterns in CLASS B (Fig 4E). This might link to the high\nexpression of DNA methyltransferase family members (DNMT1, DNMT3A, and DNMT3B) in this subtype (Fig 4F).\nMachine learning-based diagnostic models for HCC subtypes and identify potential therapeutic targets\nIn China, liver lesion biopsy is essential for HCC treatment and prognosis determination. To translate research findings into clinical\napplications, we used machine learning to identify diagnostic markers for HCC subtypes and built corresponding diagnostic models\n(Fig 5A). Specifically, we first selected differentially expressed genes in the subtypes (p.adjust < 0.001,|log2FC| > 1). Next, we\nfiltered out genes with overlapping expression patterns between the subtypes and utilized elastic net regularization to identify\ninformative genes for classification. From the elastic net results, we identified 10 significant genes for classification. Using a support\nvector machine (SVM), we created the SVM_10 classification model with 70% of the data for training and 30% for validation.\nSVM_10 achieved 87% accuracy on the validation set, with high AUC values confirming its reliability and effectiveness (Fig 5B–5D\nand Table 2).\nFig 5. Machine learning classifier for HCC subtypes.\n(A) Workflow of the subtype SVM classifier. The process of selecting subtype-specific genes followed the differential analysis\nprocedure described in the Methods, with genes retained based on criteria of p.adjust < 0.001 and |log2FC| > 1. The mean\ninterval was defined as [μ−σ, μ+σ], where μ represents the average expression of the gene in the patient cohort, and σ\nrepresents the standard deviation of gene expression in the patient cohort. (B) Predictive results of the SVM_10 model on the\nvalidation set. (C) ROC curve of the SVM_10 model on the training set. (D) ROC curve of the SVM_10 model on the\nvalidation set. (E) Expression distribution of the 10 classification genes across these two subtypes. (F) Heatmap of the\nexpression levels of the 10 classification genes in different subtypes and normal samples.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g005\nTable 2. Classification results of subtype classifiers in the validation set.\nhttps://doi.org/10.1371/journal.pcbi.1012113.t002\nSeveral studies provided evidence linking the dysregulated expression of these 10 classifier genes to adverse outcomes in HCC,\nincluding poor prognosis, increased proliferation, metastasis, and recurrence [43–50]. Notably, these genes showed significant\nexpression differences between different subtypes and normal samples (Fig 5E and 5F), indicating their crucial involvement in HCC\nprogression and potential as therapeutic targets. Among these genes, TTK has related drug information in the DrugBank database\n(https://go.drugbank.com/), indicating its potential as a therapeutic marker for HCC subtypes.\nTo assess the independent classification performance of TTK, we developed an additional SVM classifier, SVM_TTK, using TTK\nexpression values. SVM_TTK demonstrated good classification performance, with 84% accuracy on the validation set (S7A Fig and\nTable 2). It also showed high AUC values on the validation and training sets (0.91 and 0.88, respectively, S7B and S7C Fig).\nSVM_TTK effectively classified patients from the Fudan cohort (n = 225, GEO accession number: GSE14520) [42] into two\nsubtypes with significant survival differences (S8A Fig). These subtypes showed distinct TTK gene expression patterns (Mann-\nWhitney U test, S8B Fig), highlighting TTK as a potential diagnostic marker for HCC.\nClinical and prognostic characteristics of HCC subtypes with interactive survival prediction Tool\nThe two subtypes showed significant differences in clinical characteristics such as gender, age, and tumor stage (chi-square test,\nS4 Table). Patients with CLASS A had a better prognosis and mainly exhibited lower alpha-fetoprotein (AFP) expression levels in\nearly-stage samples (Mann-Whitney U test). On the other hand, CLASS B had a higher proportion of mid to late-stage samples,\nhigher AFP expression levels, and a higher frequency of viral infection.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n5/16\n\n--- Page 6 ---\nUnivariate and multivariate Cox regression analyses confirmed that the subtypes were independent prognostic factors for HCC\npatients (Fig 6A and 6B). Considering the lack of relatively comprehensive clinical information in the other two cohorts, we\ndeveloped a multivariable prognostic model using the TCGA cohort. This model integrates age, viral infection, tumor staging, and\nthe probability of CLASS B output by a machine learning diagnostic tool (Fig 6C). The C-index for this model is 0.698 (95% CI =\n0.671 ~ 0.723), and the IBS is 0.166. Illustrated through a calibration plot, we demonstrate the model’s outstanding accuracy and\npredictive performance concerning 1, 3, and 5-year survival rates (Fig 6D and 6E).\nFig 6. Construction and Evaluation of Clinical Prognostic Model.\n(A) Univariate Cox analysis of clinical features and subtypes in TCGA cohort. (B) Multivariate Cox analysis of clinical features\nand subtypes in TCGA cohort. (C) Nomogram model predicting HCC patients’ prognosis. (D) Calibration plot showing 1-, 3-,\nand 5-year survival probabilities for the nomogram model. (E) ROC curve evaluating predictive performance of the nomogram\nmodel in TCGA cohort.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g006\nFurthermore, decision curves indicated the superiority of the nomogram model over independent prognostic models using other\npredictors (S9 Fig). Additionally, we validated the model using the GSE14520 cohort [42] and achieved comparable AUC values\n(S8C Fig). To facilitate clinical application, we developed a user-friendly website allowing input of relevant information to\nautomatically generate survival plots and probabilities (https://mike-wang-bjut.shinyapps.io/DynNomapp_HCC_Sutypes/).\nAnalyzing subtype differences at the single-cell level\nTo understand the differences between these two subtypes at a higher resolution, we downloaded scRNA-seq data from 10 primary\nHCC patients from GSE149614 [51] and performed cellular-level QC. Pseudo-bulk data was created by summing gene expression\nacross sample cells. After standardization, the data was classified into two subtypes using the SVM_10 model (Fig 7A).\nFig 7. Single-cell analysis of HCC subtypes.\n(A) SVM_10 assigned subtypes to 10 primary HCC samples from GSE149614. (B) UMAP plot showing 21 cell clusters. (C)\nCell type annotation in different subtypes using marker genes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g007\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n6/16\n\n--- Page 7 ---\nUsing the UMAP method, we identified 21 cell clusters, and each cluster was annotated through the gene enrichment analysis\nmethod (Fig 7B). The single-cell-level results strongly agreed with our bulk-level findings, indicating that CLASS B had enriched\nimmune cells and cancer stem cells compared to CLASS A (Fig 7C). Moreover, T cells from CLASS B exhibited higher expression\nlevels of four immune checkpoint genes, consistent with their differential expression at the bulk level (S10 Fig). These findings\nfurther prove that the CLASS B subtype may have a higher stemness phenotype and could be more responsive to targeted immune\ncheckpoint therapies in HCC.\nDrug sensitivity differences across subtypes\nFinally, to investigate the differences in drug sensitivity across subtypes, we applied the SVM_10 model to the LIMORE dataset,\ncontaining the mutation, RNA, and drug response data for 81 HCC cell lines [52]. The SVM_10 model successfully classified the\ncell lines into two subtypes, similar to the clinical patient subtypes (Fig 8A and 8B). CLASS B showed stronger sensitivity to cell\nproliferation inhibitors, such as Temsirolimus, Camptothecin, and BX-912, possibly related to its strong proliferative characteristics.\n(Fig 8C).\nFig 8. The drug sensitivity analysis of HCC subtypes.\n(A) The SVM_10 model assigned subtypes to 81 HCC cell lines from the LIMORE dataset. (B) KEGG enrichment analysis of\nthe two subtypes. (C) Box plots illustrating drugs with differential activity area in the two subtypes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.g008\nDiscussion\nThis study expanded HCC driver genes using gene families and identified protein domains with significant mutation burden, through\npermutation testing [12]. From the results, some of these domains were associated with classical cancer mutation events, such as\nRTK and PI3K/AKT signaling pathways, along with the SET domain, known for its methyltransferase activity crucial for maintaining\nthe tumor-suppressive function of genes [53]. Moreover, the zf-H2C2_2 domain in zinc finger transcription factors also showed high\nmutation frequency and entropy, potentially leading to widespread transcriptional dysregulation in tumors and conferring a selective\ngrowth advantage to the tumor [10].\nTo overcome the discreteness inherent to mutation data for cancer stratification, we employed the NBS algorithm [27] to transform it\ninto continuous features. These features were subsequently integrated with gene expression data using the SNF algorithm [28] for\nclustering analysis. The results demonstrated that SNF effectively captured the smoothed mutation features from NBS and utilized\nthem for clustering. Through consensus clustering, we classified HCC into two subtypes, CLASS A and CLASS B, with significant\ndifferences in survival, with CLASS B displaying a lower survival probability. Moreover, compared with previous studies, our subtype\nmodel exhibits a higher value in clinical prediction.\nInterestingly, only driver genes showed poor stability in clustering and survival differences compared to the stratification results\nobtained through gene family expansion. This may be attributed to insufficiently utilizing the entire stratification algorithm’s\nextensive protein-protein interaction network information when focusing solely on driver genes. In the stratification algorithm, gene\nfamily expansion allowed for a more comprehensive consideration of family members associated with driver genes. This approach\nintroduced more relevant information, aiding in the revelation of complex molecular relationships and regulatory mechanisms. By\nincorporating family members into consideration, our understanding of the overall biological network was enhanced, resulting in\nmore biologically reasonable stratification outcomes.\nSubsequently, we conducted further analysis of the two subtypes. GSVA results indicated that CLASS A exhibited prominent\nmetabolic features enriched in pathways such as organic acid metabolism, redox reactions, fatty acid metabolism, and glycolysis.\nConversely, CLASS B was enriched in proliferative pathways, including mitosis and the cell cycle. Interestingly, we did not explicitly\nemphasize metabolism-related genes in the classifier genes. Therefore, the heightened metabolic features in CLASS A suggest that\naberrant changes in specific metabolic genes may contribute to the progression of HCC.\nMoreover, the increased dependency on these metabolic pathways may lead to metabolic vulnerability, implying the potential\ntherapeutic efficacy of inhibiting these pathways in treating CLASS A tumors [54]. Meanwhile, CLASS B exhibited higher immune\ncell abundance and overexpression of immune checkpoint molecules such as CTLA4 and HCVAR2, indicating its potential\nsuitability for immune checkpoint inhibitor therapy. Additionally, CLASS B had higher stemness scores, implying a more active\npopulation of cancer stem cells and increased tumor cell dedifferentiation [38]. These findings were confirmed in subsequent single-\ncell analysis.\nNext, we compared genomic differences between the subtypes and observed significant chromosomal instability in CLASS B,\ncharacterized by deletions in chromosomes 4p, 4q, 13q, 16p, 16q, and 17p. These deletions are frequently found in clinically\nadvanced, poorly differentiated, large, and metastatic HCC cases [55,56]. Additionally, our subtypes demonstrated a high degree of\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n7/16\n\n--- Page 8 ---\nconsistency with other subtypes. For example, CLASS A was enriched in several high metabolic subtypes with a better prognosis,\nsuch as Hosdia S3, iHCC1, and hALDH2. Additionally, CLASS B was enriched in low metabolic flux, high TP53 mutation, and highly\nproliferative subtypes, such as Hosdia S1, Hosdia S2, and iHCC3. In addition, referring to the findings of Benfeitas et al. [30], these\ntwo subtypes may employ different mechanisms to resist reactive oxygen species (ROS). Notably, during the subtype stratification\nof HCC, we incorporated mutation data. We observed significant differences in the mutation frequencies of classical oncogenes,\nsuch as RB1, ALB, APOB, and NFE2L2, among others, between the two distinct subtypes, except for TP53 and CTNNB1. This\ndifference was not significant among other subtypes (p > 0.05).\nRB1, a crucial tumor suppressor gene, regulates the cell cycle by inhibiting E2F transcription factors and cyclin-dependent kinases\nduring G1 to S phase transition [57,58]. RB1 mutations were predominantly found in CLASS B samples (Fisher’s exact test, p =\n1.39×10–8), explaining the heightened proliferative features in CLASS B. The Wnt-β-catenin signaling pathway is frequently\nactivated in HCC [59]. Aggressive HCC subtypes, unlike well-prognosed ones, enhance the Wnt pathway by regulating intracellular\nfree β-catenin through TGF-β overexpression [31]. In CLASS B, we observed the RGS domain primarily occurring in AXIN1\n(Fisher’s exact test, p = 0.012632, S11 Fig and S7 Table). This domain binds to APC protein, contributing to β-catenin degradation\n[60]. Missense mutations in the RGS domain of AXIN1 promote AXIN1 aggregation, leading to impaired β-catenin degradation [61].\nThus, in this aggressive subtype, gene mutations may indirectly or directly affect β-catenin degradation, resulting in Wnt pathway\nactivation.\nFinally, we developed a 10-gene SVM subtypes classifier between the subtypes. An intriguing finding was made regarding TTK,\none of the classifier genes with relevant drug records in Drugbank. TTK itself demonstrated excellent performance in subtype\nclassification. TTK, also known as Mps1, recruits other SAC proteins to unattached kinetochores during prophase to activate SAC-\nrelated arrest [62]. Inhibiting TTK activity leads to premature chromosome segregation, severe chromosomal missegregation,\naneuploidy, and cell death [63]. TTK is overexpressed in various human tumors, including HCC. Studies indicate its role in\npromoting HCC cell malignancy [64]. TTK inhibitors like BOS172722 [65] and CFI-402257 [66] have shown promise in cancer\ntreatment, making TTK a potential independent diagnostic biomarker and therapeutic target for specific HCC subtypes.\nAlthough we have identified two HCC subtypes with distinct survival and biological characteristics using driver genes and their\nfamily members, there are some limitations. Firstly, the lack of patient cohorts with comprehensive multi-omics and clinical data\nrestricts the validation of the subtype models across diverse populations. Secondly, this study needs to include relevant in vivo/in\nvitro experiments to demonstrate the therapeutic effect of relevant inhibitors on specific subtypes.\nConclusion\nIn conclusion, we successfully constructed two HCC subtypes with significant differences in survival. We have delved into the\nbiological disparities underlying these subtypes by integrating diverse omics data. Our tailored prognostic and classification models\ndemonstrated robustness and accurately predicted overall survival in HCC patients. Notably, TTK emerged as a crucial diagnostic\nbiomarker and potential therapeutic target for specific HCC subtypes. Our study offers a fresh outlook on HCC subtypes, enhancing\nour understanding of the disease’s pathogenesis and potential therapeutic strategies.\nMethods\nData sources and preprocessing\nThis study included a cohort of 893 patients with primary tumors from three HCC datasets, namely TCGA_LIHC, ICGC_JP, and\nICGC_FR. Among these, 690 patients (S8 Table) were retained for subsequent analysis as they had both RNA-seq and mutation\ndata available.\nSomatic mutations, miRNA expression, HM450 methylation, copy number variants, and clinical information for the TCGA_LIHC\ncohort were obtained from the GDC Data Portal(https://portal.gdc.cancer.gov/), mRNA-Seq data were obtained from the\nTCGAxGETx combined dataset from UCSC Xena (http://xena.ucsc.edu/). Somatic mutations, mRNA-Seq, and clinical data for\nICGC_JP and ICGC_FR cohorts were obtained from the ICGC database (https://dcc.icgc.org/).\nWe standardized them into TPM values to ensure consistency and comparability of RNA-seq data across the three cohorts. We\nmitigated batch effects using the \"ComBat\" function within the R package \"sva\" [67]. After amalgamating these cohorts into a single\nmerged cohort, we filtered out genes exhibiting TPM values ≤ 0 in over 30% of the samples. This preliminary data processing step\neffectively reduced noise, ensuring data quality by retaining approximately 17,660 genes for subsequent analyses.\nMethylation probes were preprocessed and normalized using the R package \"ChAMP\" [68]. We removed null rate probes in more\nthan 30% of methylation and miRNA data samples, using KNN (K = 15) interpolation with the R package \"impute\".\nFor somatic mutation data, we retained only non-silent mutations in the coding region.\nProtein domain with significant mutation burden\nBefore the domain mutation analysis, we further processed the somatic mutation data. Samples with tumor mutational burden\n(TMB) values >30 mutations/Mb were excluded to eliminate the influence of hypermutated samples. We calculated TMB as the\ncount of non-silent mutations within coding region exons divided by 38 Mb. We also removed MAF entries with non-point mutations,\namino acid change positions larger than the protein length, and the reference amino acids not aligning with the protein reference\nsequence.\nA permutation test was used to identify protein domains with significant mutation burden, as described by Miller et al. [12] In this\ntest, we assumed that all amino acids within the protein have an equal chance of mutation. The permutation test assessed whether\nthe observed values significantly deviated from the empirical distribution obtained from random mutations in the gene. The P-value\nafter i permutations was defined as:\nTo quantify the information about the distribution of the mutation burden within a specific domain among its gene members, we\ncomputed the entropy value (S). The entropy is defined as:\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n8/16\n\n--- Page 9 ---\nIn information entropy theory, the entropy value reaches its maximum when the distribution is uniform. Thus, the entropy value is\nclose to 1 if mutations in the domain are uniformly distributed across the family genes. Conversely, if the mutations are not\nuniformly distributed, the entropy value will be closer to 0.\nDifferential analysis\nWe used the Mann-Whitney U test for miRNA and mRNA data to find genes with expression differences between different groups.\nWe corrected the p-values using the Benjamin-Hochberg method. We define differentially expressed miRNA (DMi) and differential\ngene (DEG) for those with | Log2FC |>1 and p.adjust<0.05. Additionally, we excluded DMi and DEG with mean expression levels\nless than 1 in both the test and control groups.\nDifferential methylation probes (DMPs) were identified using the R package \"ChAMP\" for DNA methylation data, considering probes\nwith |Δβ| > 0.2 and p.adjust < 0.05 as DMPs.\nFor copy number data, we employed the GISTIC2.0 function available in GenePattern(https://www.genepattern.org/) to identify\nchromosomal regions and genes exhibiting significant copy number variations.\nDefine driver-dysregulated gene\nWe defined driver-dysregulated genes (DDGs) as genes that exhibit transcriptional dysregulation due to alterations in DNA\nmethylation, miRNA expression, or copy number variation. To comprehensively define DDGs, we further divide DMPs into three\ngroups: distal enhancers, promoters (TSS1500, TSS200, 5’UTR, and 1stExon), and gene body. Here, TSS1500 refers to the region\n200–1500 bases upstream of the transcriptional start site (TSS), while TSS200 represents the region 0–200 bases upstream of the\nTSS. The R packages \"ELMERv.2\" [69] and \"ChAMP\" were used to identify distal enhancer probes, promoter and gene body\nmethylation probes, and their target genes. For DMi, we used the R package \"multiMiR\" [70] to obtain target genes based on\nexperimental or computational predictions. Genes with a copy number gain or loss ratio greater than 20% were categorized as gain\nor loss groups, respectively (CNV levels greater than 1 for gain and less than -1 for loss). The association between DMPs, DMi,\nCNV, and DEG was analyzed using Spearman correlation (threshold: |ρ| > 0.2, p-value after Benjamin Hochberg correction < 0.05).\nVenn diagrams were used for result filtering to illustrate the relationships between different omics layers and the transcriptome.\nSubtype recognition model\nIn this paper, we aimed to construct HCC subtypes using somatic mutation data and mRNA expression data. First, we converted\nthe mutation MAF file into a binary matrix, where 0 represents no mutation, and 1 indicates a mutation presence in that sample. To\nsmooth the mutation data, we used pyNBS (https://github.com/idekerlab/pyNBS) [71], a Python version of the Network-based\nstratification (NBS) algorithm [27]. This algorithm employs network propagation to smooth the mutation signals, enhancing their\nclassification capabilities like other continuous features. The following formula can represent the process of network propagation:\nF represents the patient mutation matrix, A represents the normalized adjacency matrix of the gene interaction network, and α is\nan adjustment parameter that controls the distance allowed for mutation signals to spread through the network during propagation.\nThe propagation function iterates until convergence (determined by the matrix norm of F\n- F < 1×10\n), with α set to 0.7 following\nrecommendations from the references for smoothing the mutation data.\nThe gene expression data underwent log (x+1) transformation and Z-score normalization. Next, we integrated the smoothed\nmutation data with the gene expression data to form a similarity matrix W using the R package \"SNFtool\". For integration, we used\n20 nearest neighbors, set the variance of the local model to 0.5, and performed 20 iterations of the diffusion process.\nWe conducted consensus clustering using the similarity matrix W with the R package \"ConsensusClusterPlus.\" We set clustering\nparameters as follows: maximum clusters = 6, iterations = 5000, item sample proportion = 0.8, distance metric = ’spearman’, and\nclustering algorithm = ’hc’. We evaluated clustering quality using the silhouette coefficient (Sil), defined as follows\nHere, a(i) represents the average distance between vector i and all other points within the same cluster. In contrast, b(i) represents\nthe minimum average distance between vector i and all points in a cluster that does not include it.\nGene enrichment analysis\nThe R package \"clusterProfiler\" [72] performs hypergeometric distribution tests on the annotations of specific gene sets in different\ndatabases.\nTo convert gene expression data into scores for particular biological processes, we utilized the R packages \"GSVA\" [73] and\n\"msigdbr.\" Subsequently, a Student’s t-test was employed to identify significant differences in biological processes among different\nsubtypes. A differential biological process was defined as |ΔGSVA score| > 0.2 and Benjamini-Hochberg adjusted p < 0.05.\nCIN ratio\nBased on previous research, we assessed chromosomal instability (CIN) across different subtypes using the CIN ratio [74,75]. To\ncalculate the CIN ratio for each tumor sample, we extracted the \"broad_values_by_arm.txt\" file from the GISTIC2.0 results. In this\nfile, CNV scores exceeding 0.1 or falling below -0.1 were regarded as alterations. The CIN ratio was defined as:\n0\nt+1\nt\n−6\n2\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n9/16\n\n--- Page 10 ---\nNomogram model of HCC subtypes\nWe used the R package \"survival\" to construct univariate and multivariate Cox prognostic models. The \"forestplot\" package was\nemployed to generate forest plots, while the \"rms\" package facilitated the creation of a nomogram for the multivariate model. To\nassess the performance of the nomogram model, we employed calibration plots, decision curve analysis, and receiver operating\ncharacteristic (ROC) curves, leveraging the R packages \"rms,\" \"dcurves,\" and \"timeROC\" respectively. Additionally, the R package\n\"DynNom\" wasutilized to develop a dynamic nomogram model and an interactive webpage.\nSingle-cell data processing\nWe downloaded the HCC single-cell RNA sequencing (scRNA-seq) data from the GEO database (GEO accession: GSE149614\n[51]). For further analysis, we selected 10 primary tumor samples. Quality control (QC) was performed using the R package\n\"Seurat\" [76], involving cell-level QC and gene-level QC. Specifically, cells with UMIs greater than 500, expressing genes between\n500 and 8000, and mitochondrial content less than 10% were retained. We kept genes with expression data in at least 10 cells for\ngene-level QC. Subsequently, 22,298 genes and 31,490 cells were used for subsequent analyses.\nSingle-cell data dimensionality reduction clustering\nWe processed the scRNA-seq data using the R package \"Seurat.\" Firstly, the \"NormalizeData\" function was applied for background\ncorrection and normalization. We employed the \"FindVariableFeatures\" function (selection method = ’vst’, ’ x-axis cut off = (0.0125,\n3), y-axis cut off > 0.5) to identify the top 2000 highly variable genes. Scaling of the data was carried out with the \"ScaleData\"\nfunction, excluding mitochondrial contamination heterogeneity. Next, we reduced data dimensionality with the \"RunPCA\" function,\nselecting the top 30 principal components based on the \"ElbowPlot\" analysis. Cell clustering was conducted with a resolution of 0.2\nusing the \"FindClusters\" function, and cell clusters were visualized with the \"RunUMAP\" and \"DimPlot\" functions. Significant marker\ngenes in different clusters were identified using the \"FindAllMarkers\" function. To annotate the cell clusters, we adopted a statistical-\nbased approach by performing enrichment analysis on the marker genes using the \"clusterProfiler\" R package. We downloaded the\nHuman cell markers dataset from CellMarker (http://biocc.hrbmu.edu.cn/CellMarker).\nSubtype machine learning model\nAll machine learning classifiers were constructed using the \"scikit-learn\" package (version 1.0.2) in Python (version 3.9.12). Cross-\nvalidation and grid search were used to obtain the best hyperparameters for the model. To assess the performance of the\nclassification results, we employed Confusion matrices, ROC curves, accuracy (ACC), sensitivity (SEN), specificity (SPE), and\nMatthews correlation coefficient (MCC). These evaluation metrics are defined as follows:\nHere, TP denotes the count of true positives, TN denotes the count of true negatives, FP denotes the count of false positives, and\nFN denotes the count of false negatives. The machine learning model can be found at https://github.com/Mike-\nW29/SVM_model_for_HCC_subtype.\nStatistic analysis\nStatistical analyses for this paper were conducted using R (version: 4.2.1). Kaplan-Meier curves were evaluated with the log-rank\ntest to determine significance. The chi-square and Fisher’s exact tests explored associations between clinical characteristics in\ndifferent groups. Unless otherwise specified, statistical significance was defined as p-values < 0.05 (two-tailed). In the paper,\nsignificance levels were denoted as * (p < 0.05), ** (p < 0.01), *** (p < 0.001), **** (p < 0.0001), and \"na\" indicated no statistical\ndifference.\nSupporting information\nS1 Fig. Identification of driver gene-related subtypes in the TCGA cohort.\n(A) Silhouette coefficients for different values of k. (B) Silhouette plot for k = 2. (C) Consensus matrix heatmap defining two\nsubtypes. (D) Five-year survival curves for the two subtypes, with CLASS A represented in red and CLASS B in blue.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s001\n(TIF)\nS2 Fig. The Sankey diagrams depicting the associations between the subtypes identified in our study and those from previous research studies.\n(A) Hoshida 3-class subtypes. (B) Bidkhori 3-class subtypes. (C) Benfeitas 2-class subtypes. (D) TCGA 3-class subtypes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s002\n(TIF)\nS3 Fig. C-index analysis comparing the subtypes identified in this study with previous research.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s003\n(TIF)\nS4 Fig. XCELL analysis results for the two subtypes.\n(A) Analysis of different cell abundances between the two subtypes using XCELL. The statistical significance of the differences was\nassessed using the Mann-Whitney U test. (B) Analysis of different immune scores, stromal scores, and microenvironment scores\nbetween the two subtypes using XCELL. The statistical significance of the differences was assessed using the Mann-Whitney U\ntest.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s004\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n10/16\n\n--- Page 11 ---\n1.\nView Article\nPubMed/NCBI\nGoogle Scholar\n(TIF)\nS5 Fig. Box plots depicting the expression of KRT19 in the two subtypes.\nStatistical significance of the differences was assessed using the Mann-Whitney U test to determine if these differences were\nstatistically significant.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s005\n(TIF)\nS6 Fig. Box plot showing the TMB in each subtype.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s006\n(TIF)\nS7 Fig. The classification results of the HCC subtypes using the SVM_TTK model.\n(A) Predicted results of the SVM_TTK model on the validation set. (B) ROC curve of the SVM_TTK model in the training set. (C)\nROC curve of the SVM_TTK model in the validation set.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s007\n(TIF)\nS8 Fig. The validation of the classification and prognostic models.\n(A) Survival analysis of the GSE14520 cohort based on the classification results using the SVM_TKK model. (B) Expression levels\nof TTK in the two subtypes of the GSE14520 cohort, with statistical significance determined by the Mann-Whitney U test (C) ROC\ncurves depicting the predictive results of the prognostic model for 1-, 3-, and 5-year survival probabilities in the GSE14520 cohort.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s008\n(TIF)\nS9 Fig. The decision curve analysis for the prognostic model of the subtypes.\n(A) Decision curve analysis for 1-year survival. (B) Decision curve analysis for 3-year survival. (C) Decision curve analysis for 5-\nyear survival.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s009\n(TIF)\nS10 Fig. The single-cell expression profiles of four immune checkpoint genes that exhibit differential expression at the bulk level.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s010\n(TIF)\nS11 Fig. Oncoplot illustrating the specific differences in protein domain mutation frequencies between the two subtypes (chi-square test).\nhttps://doi.org/10.1371/journal.pcbi.1012113.s011\n(TIF)\nS1 Table. Driver gene and their family number.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s012\n(XLS)\nS2 Table. Protein domains with significant mutations.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s013\n(XLS)\nS3 Table. List of stratification genes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s014\n(XLS)\nS4 Table. Differential clinical features between the two subtypes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s015\n(XLS)\nS5 Table. HCC subtypes from other studies used in this paper.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s016\n(XLS)\nS6 Table. Differential enrichment analysis of two subtypes in REACTOME gene sets using GSVA.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s017\n(XLS)\nS7 Table. Protein domains with specific differences in mutation frequencies between the two subtypes.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s018\n(XLS)\nS8 Table. Description of the patient cohort.\nhttps://doi.org/10.1371/journal.pcbi.1012113.s019\n(XLS)\nAcknowledgments\nWe thank Dr Jianming Zeng (University of Macau), and all the members of his bioinformatics team, biotrainee, for generously\nsharing their experience and codes.\nReferences\nSung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and\nMortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209–49. Epub 2021/02/05. pmid:33538338.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n11/16\n\n--- Page 12 ---\n2.\nView Article\nPubMed/NCBI\nGoogle Scholar\n3.\nView Article\nPubMed/NCBI\nGoogle Scholar\n4.\nView Article\nPubMed/NCBI\nGoogle Scholar\n5.\nView Article\nPubMed/NCBI\nGoogle Scholar\n6.\nView Article\nPubMed/NCBI\nGoogle Scholar\n7.\nView Article\nPubMed/NCBI\nGoogle Scholar\n8.\nView Article\nPubMed/NCBI\nGoogle Scholar\n9.\nView Article\nPubMed/NCBI\nGoogle Scholar\n10.\nView Article\nPubMed/NCBI\nGoogle Scholar\n11.\nView Article\nPubMed/NCBI\nGoogle Scholar\n12.\nView Article\nPubMed/NCBI\nGoogle Scholar\n13.\nView Article\nPubMed/NCBI\nGoogle Scholar\n14.\nView Article\nPubMed/NCBI\nGoogle Scholar\n15.\nView Article\nPubMed/NCBI\nGoogle Scholar\n16.\nView Article\nPubMed/NCBI\nGoogle Scholar\n17.\nView Article\nPubMed/NCBI\nGoogle Scholar\n18.\nView Article\nPubMed/NCBI\nGoogle Scholar\n19.\nView Article\nPubMed/NCBI\nGoogle Scholar\nBertuccio P, Turati F, Carioli G, Rodriguez T, La Vecchia C, Malvezzi M, et al. Global trends and predictions in hepatocellular carcinoma mortality. J\nHepatol. 2017;67(2):302–9. Epub 2017/03/25. pmid:28336466.\nWu Y, Liu Z, Xu X. Molecular subtyping of hepatocellular carcinoma: A step toward precision medicine. Cancer Commun (Lond). 2020;40(12):681–93.\nEpub 2020/12/09. pmid:33290597.\nNicora G, Vitali F, Dagliati A, Geifman N, Bellazzi R. Integrated Multi-Omics Analyses in Oncology: A Review of Machine Learning Methods and Tools.\nFront Oncol. 2020;10:1030. Epub 2020/07/23. pmid:32695678.\nSia D, Villanueva A, Friedman SL, Llovet JM. Liver Cancer Cell of Origin, Molecular Class, and Effects on Patient Prognosis. Gastroenterology.\n2017;152(4):745–61. Epub 2017/01/04. pmid:28043904.\nZhang Q, Lou Y, Yang J, Wang J, Feng J, Zhao Y, et al. Integrated multiomic analysis reveals comprehensive tumour heterogeneity and novel\nimmunophenotypic classification in hepatocellular carcinomas. Gut. 2019;68(11):2019–31. Epub 2019/06/23. pmid:31227589.\nPoirion OB, Jing Z, Chaudhary K, Huang S, Garmire LX. DeepProg: an ensemble of deep-learning and machine-learning models for prognosis prediction\nusing multi-omics data. Genome Med. 2021;13(1):112. Epub 2021/07/16. pmid:34261540.\nWaks Z, Weissbrod O, Carmeli B, Norel R, Utro F, Goldschmidt Y. Driver gene classification reveals a substantial overrepresentation of tumor suppressors\namong very large chromatin-regulating proteins. Sci Rep. 2016;6:38988. Epub 2016/12/23. pmid:28008934.\nLlovet JM, Kelley RK, Villanueva A, Singal AG, Pikarsky E, Roayaie S, et al. Hepatocellular carcinoma. Nat Rev Dis Primers. 2021;7(1):6. Epub\n2021/01/23. pmid:33479224.\nMunro D, Ghersi D, Singh M. Two critical positions in zinc finger domains are heavily mutated in three human cancer types. PLoS Comput Biol.\n2018;14(6):e1006290. Epub 2018/06/29. pmid:29953437.\nDavies H, Hunter C, Smith R, Stephens P, Greenman C, Bignell G, et al. Somatic mutations of the protein kinase gene family in human lung cancer.\nCancer Res. 2005;65(17):7591–5. Epub 2005/09/06. pmid:16140923.\nMiller ML, Reznik E, Gauthier NP, Aksoy BA, Korkut A, Gao J, et al. Pan-Cancer Analysis of Mutation Hotspots in Protein Domains. Cell Syst.\n2015;1(3):197–209. Epub 2016/05/03. pmid:27135912.\nBailey MH, Tokheim C, Porta-Pardo E, Sengupta S, Bertrand D, Weerasinghe A, et al. Comprehensive Characterization of Cancer Driver Genes and\nMutations. Cell. 2018;173(2):371–85 e18. Epub 2018/04/07. pmid:29625053.\nMartinez-Jimenez F, Muinos F, Sentis I, Deu-Pons J, Reyes-Salazar I, Arnedo-Pac C, et al. A compendium of mutational cancer driver genes. Nat Rev\nCancer. 2020;20(10):555–72. Epub 2020/08/12. pmid:32778778.\nFujimoto A, Furuta M, Totoki Y, Tsunoda T, Kato M, Shiraishi Y, et al. Whole-genome mutational landscape and characterization of noncoding and\nstructural mutations in liver cancer. Nat Genet. 2016;48(5):500–9. Epub 2016/04/12. pmid:27064257.\nYamamoto-Furusho JK, Fonseca-Camarillo G, Furuzawa-Carballeda J, Sarmiento-Aguilar A, Barreto-Zuniga R, Martinez-Benitez B, et al. Caspase\nrecruitment domain (CARD) family (CARD9, CARD10, CARD11, CARD14 and CARD15) are increased during active inflammation in patients with\ninflammatory bowel disease. J Inflamm (Lond). 2018;15:13. Epub 2018/07/17. pmid:30008619.\nPino I, Pio R, Toledo G, Zabalegui N, Vicent S, Rey N, et al. Altered patterns of expression of members of the heterogeneous nuclear ribonucleoprotein\n(hnRNP) family in lung cancer. Lung Cancer. 2003;41(2):131–43. Epub 2003/07/23. pmid:12871776.\nKatoh Y, Katoh M. Comparative integromics on FAT1, FAT2, FAT3 and FAT4. Int J Mol Med. 2006;18(3):523–8. Epub 2006/07/26. pmid:16865240.\nKatoh M. Function and cancer genomics of FAT family genes (review). Int J Oncol. 2012;41(6):1913–8. Epub 2012/10/19. pmid:23076869.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n12/16\n\n--- Page 13 ---\n20.\nView Article\nPubMed/NCBI\nGoogle Scholar\n21.\nView Article\nPubMed/NCBI\nGoogle Scholar\n22.\nView Article\nPubMed/NCBI\nGoogle Scholar\n23.\nView Article\nPubMed/NCBI\nGoogle Scholar\n24.\nView Article\nPubMed/NCBI\nGoogle Scholar\n25.\nView Article\nPubMed/NCBI\nGoogle Scholar\n26.\nView Article\nPubMed/NCBI\nGoogle Scholar\n27.\nView Article\nPubMed/NCBI\nGoogle Scholar\n28.\nView Article\nPubMed/NCBI\nGoogle Scholar\n29.\nView Article\nPubMed/NCBI\nGoogle Scholar\n30.\nView Article\nPubMed/NCBI\nGoogle Scholar\n31.\nView Article\nPubMed/NCBI\nGoogle Scholar\n32.\nView Article\nPubMed/NCBI\nGoogle Scholar\n33.\nView Article\nPubMed/NCBI\nGoogle Scholar\n34.\nView Article\nPubMed/NCBI\nGoogle Scholar\n35.\nView Article\nPubMed/NCBI\nGoogle Scholar\n36.\nView Article\nPubMed/NCBI\nGoogle Scholar\n37.\nView Article\nPubMed/NCBI\nGoogle Scholar\nLu Y, Liu Z, Wang W, Chen X, Zhou X, Fu W. Expression Signature of the AT-Rich Interactive Domain Gene Family Identified in Digestive Cancer. Front\nMed (Lausanne). 2021;8:775357. Epub 2022/02/08. pmid:35127746.\nZhang H, Cao X, Wang J, Li Q, Zhao Y, Jin X. LZTR1: A promising adaptor of the CUL3 family. Oncol Lett. 2021;22(1):564. Epub 2021/06/12.\npmid:34113392.\nTokheim C, Wang X, Timms RT, Zhang B, Mena EL, Wang B, et al. Systematic characterization of mutations altering protein degradation in human\ncancers. Mol Cell. 2021;81(6):1292–308 e11. Epub 2021/02/11. pmid:33567269.\nHou C, Li Y, Wang M, Wu H, Li T. Systematic prediction of degrons and E3 ubiquitin ligase binding via deep learning. BMC Biol. 2022;20(1):162. Epub\n2022/07/15. pmid:35836176.\nCsepregi A, Ebert MP, Rocken C, Schneider-Stock R, Hoffmann J, Schulz HU, et al. Promoter methylation of CDKN2A and lack of p16 expression\ncharacterize patients with hepatocellular carcinoma. BMC Cancer. 2010;10:317. Epub 2010/06/24. pmid:20569442.\nLiu H, Dong H, Robertson K, Liu C. DNA methylation suppresses expression of the urea cycle enzyme carbamoyl phosphate synthetase 1 (CPS1) in\nhuman hepatocellular carcinoma. Am J Pathol. 2011;178(2):652–61. Epub 2011/02/02. pmid:21281797.\nSong Z, Yu Z, Chen L, Zhou Z, Zou Q, Liu Y. MicroRNA-1181 supports the growth of hepatocellular carcinoma by repressing AXIN1. Biomed\nPharmacother. 2019;119:109397. Epub 2019/09/13. pmid:31514071.\nHofree M, Shen JP, Carter H, Gross A, Ideker T. Network-based stratification of tumor mutations. Nat Methods. 2013;10(11):1108–15. Epub 2013/09/17.\npmid:24037242.\nWang B, Mezlini AM, Demir F, Fiume M, Tu Z, Brudno M, et al. Similarity network fusion for aggregating data types on a genomic scale. Nat Methods.\n2014;11(3):333–7. Epub 2014/01/28. pmid:24464287.\nKim CY, Baek S, Cha J, Yang S, Kim E, Marcotte EM, et al. HumanNet v3: an improved database of human gene networks for disease research. Nucleic\nAcids Res. 2022;50(D1):D632–D9. Epub 2021/11/09. pmid:34747468.\nBenfeitas R, Bidkhori G, Mukhopadhyay B, Klevstig M, Arif M, Zhang C, et al. Characterization of heterogeneous redox responses in hepatocellular\ncarcinoma patients using network analysis. EBioMedicine. 2019;40:471–87. Epub 2019/01/05. pmid:30606699.\nHoshida Y, Nijman SM, Kobayashi M, Chan JA, Brunet JP, Chiang DY, et al. Integrative transcriptome analysis reveals common molecular subclasses of\nhuman hepatocellular carcinoma. Cancer Res. 2009;69(18):7385–92. Epub 2009/09/03. pmid:19723656.\nBidkhori G, Benfeitas R, Klevstig M, Zhang C, Nielsen J, Uhlen M, et al. Metabolic network-based stratification of hepatocellular carcinoma reveals three\ndistinct tumor subtypes. Proc Natl Acad Sci U S A. 2018;115(50):E11874–E83. Epub 2018/11/30. pmid:30482855.\nCancer Genome Atlas Research Network. Electronic address wbe, Cancer Genome Atlas Research N. Comprehensive and Integrative Genomic\nCharacterization of Hepatocellular Carcinoma. Cell. 2017;169(7):1327–41 e23. Epub 2017/06/18. pmid:28622513.\nLi T, Fu J, Zeng Z, Cohen D, Li J, Chen Q, et al. TIMER2.0 for analysis of tumor-infiltrating immune cells. Nucleic Acids Res. 2020;48(W1):W509–W14.\nEpub 2020/05/23. pmid:32442275.\nAran D, Hu Z, Butte AJ. xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol. 2017;18(1):220. Epub 2017/11/17.\npmid:29141660 The authors declare that they have no competing interests. PUBLISHER’S NOTE: Springer Nature remains neutral with regard to\njurisdictional claims in published maps and institutional affiliations.\nHarjunpaa H, Guillerey C. TIGIT as an emerging immune checkpoint. Clin Exp Immunol. 2020;200(2):108–19. Epub 2019/12/13. pmid:31828774.\nPardoll DM. The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer. 2012;12(4):252–64. Epub 2012/03/23. pmid:22437870.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n13/16\n\n--- Page 14 ---\n38.\nView Article\nPubMed/NCBI\nGoogle Scholar\n39.\nView Article\nPubMed/NCBI\nGoogle Scholar\n40.\nView Article\nPubMed/NCBI\nGoogle Scholar\n41.\nView Article\nPubMed/NCBI\nGoogle Scholar\n42.\nView Article\nPubMed/NCBI\nGoogle Scholar\n43.\nView Article\nPubMed/NCBI\nGoogle Scholar\n44.\nView Article\nPubMed/NCBI\nGoogle Scholar\n45.\nView Article\nPubMed/NCBI\nGoogle Scholar\n46.\nView Article\nPubMed/NCBI\nGoogle Scholar\n47.\nView Article\nPubMed/NCBI\nGoogle Scholar\n48.\nView Article\nPubMed/NCBI\nGoogle Scholar\n49.\nView Article\nPubMed/NCBI\nGoogle Scholar\n50.\nView Article\nPubMed/NCBI\nGoogle Scholar\n51.\nView Article\nPubMed/NCBI\nGoogle Scholar\n52.\nView Article\nPubMed/NCBI\nGoogle Scholar\n53.\nView Article\nPubMed/NCBI\nGoogle Scholar\n54.\nView Article\nPubMed/NCBI\nGoogle Scholar\n55.\nView Article\nPubMed/NCBI\nGoogle Scholar\nMalta TM, Sokolov A, Gentles AJ, Burzykowski T, Poisson L, Weinstein JN, et al. Machine Learning Identifies Stemness Features Associated with\nOncogenic Dedifferentiation. Cell. 2018;173(2):338–54 e15. Epub 2018/04/07. pmid:29625051.\nGovaere O, Komuta M, Berkers J, Spee B, Janssen C, de Luca F, et al. Keratin 19: a key role player in the invasion of human hepatocellular carcinomas.\nGut. 2014;63(4):674–85. Epub 2013/08/21. pmid:23958557.\nLiu J, Ma Q, Zhang M, Wang X, Zhang D, Li W, et al. Alterations of TP53 are associated with a poor outcome for patients with hepatocellular carcinoma:\nevidence from a systematic review and meta-analysis. Eur J Cancer. 2012;48(15):2328–38. Epub 2012/03/31. pmid:22459764.\nKhalaf AM, Fuentes D, Morshid AI, Burke MR, Kaseb AO, Hassan M, et al. Role of Wnt/beta-catenin signaling in hepatocellular carcinoma, pathogenesis,\nand clinical significance. J Hepatocell Carcinoma. 2018;5:61–73. Epub 2018/07/10. pmid:29984212.\nRoessler S, Jia HL, Budhu A, Forgues M, Ye QH, Lee JS, et al. A unique metastasis gene signature enables prediction of tumor relapse in early-stage\nhepatocellular carcinoma patients. Cancer Res. 2010;70(24):10202–12. Epub 2010/12/17. pmid:21159642.\nDai B, Zhang X, Shang R, Wang J, Yang X, Zhang H, et al. Blockade of ARHGAP11A reverses malignant progress via inactivating Rac1B in\nhepatocellular carcinoma. Cell Commun Signal. 2018;16(1):99. Epub 2018/12/14. pmid:30545369.\nZhang Z, Zhang Y, Mo W. The Autophagy Related Gene CHAF1B Is a Relevant Prognostic and Diagnostic Biomarker in Hepatocellular Carcinoma. Front\nOncol. 2020;10:626175. Epub 2021/02/13. pmid:33575221.\nDang XW, Pan Q, Lin ZH, Wang HH, Li LH, Li L, et al. Overexpressed DEPDC1B contributes to the progression of hepatocellular carcinoma by CDK1.\nAging (Albany NY). 2021;13(16):20094–115. Epub 2021/05/26. pmid:34032605.\nChen J, Xia H, Zhang X, Karthik S, Pratap SV, Ooi LL, et al. ECT2 regulates the Rho/ERK signalling axis to promote early recurrence in human\nhepatocellular carcinoma. J Hepatol. 2015;62(6):1287–95. Epub 2015/01/27. pmid:25617497.\nLi S, Wu L, Zhang H, Liu X, Wang Z, Dong B, et al. GINS1 Induced Sorafenib Resistance by Promoting Cancer Stem Properties in Human Hepatocellular\nCancer Cells. Front Cell Dev Biol. 2021;9:711894. Epub 2021/08/21. pmid:34414190.\nWu X, Wang H, Lian Y, Chen L, Gu L, Wang J, et al. GTSE1 promotes cell migration and invasion by regulating EMT in hepatocellular carcinoma and is\nassociated with poor prognosis. Sci Rep. 2017;7(1):5129. Epub 2017/07/13. pmid:28698581.\nYang Y, Gao L, Chen J, Xiao W, Liu R, Kan H. Lamin B1 is a potential therapeutic target and prognostic biomarker for hepatocellular carcinoma.\nBioengineered. 2022;13(4):9211–31. Epub 2022/04/19. pmid:35436411.\nGuan Z, Cheng W, Huang D, Wei A. High MYBL2 expression and transcription regulatory activity is associated with poor overall survival in patients with\nhepatocellular carcinoma. Curr Res Transl Med. 2018;66(1):27–32. Epub 2017/12/25. pmid:29274707.\nLu Y, Yang A, Quan C, Pan Y, Zhang H, Li Y, et al. A single-cell atlas of the multicellular ecosystem of primary and metastatic hepatocellular carcinoma.\nNat Commun. 2022;13(1):4594. Epub 2022/08/07. pmid:35933472.\nQiu Z, Li H, Zhang Z, Zhu Z, He S, Wang X, et al. A Pharmacogenomic Landscape in Human Liver Cancers. Cancer Cell. 2019;36(2):179–93 e11. Epub\n2019/08/06. pmid:31378681.\nKim KC, Geng L, Huang S. Inactivation of a histone methyltransferase by mutations in human cancers. Cancer Res. 2003;63(22):7619–23. Epub\n2003/11/25. pmid:14633678.\nTenen DG, Chai L, Tan JL. Metabolic alterations and vulnerabilities in hepatocellular carcinoma. Gastroenterol Rep (Oxf). 2021;9(1):1–13. Epub\n2021/03/23. pmid:33747521.\nTsuda H, Zhang WD, Shimosato Y, Yokota J, Terada M, Sugimura T, et al. Allele loss on chromosome 16 associated with progression of human\nhepatocellular carcinoma. Proc Natl Acad Sci U S A. 1990;87(17):6791–4. Epub 1990/09/01. pmid:2168560.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n14/16\n\n--- Page 15 ---\n56.\nView Article\nPubMed/NCBI\nGoogle Scholar\n57.\nView Article\nPubMed/NCBI\nGoogle Scholar\n58.\nView Article\nPubMed/NCBI\nGoogle Scholar\n59.\nView Article\nPubMed/NCBI\nGoogle Scholar\n60.\nView Article\nPubMed/NCBI\nGoogle Scholar\n61.\nView Article\nPubMed/NCBI\nGoogle Scholar\n62.\nView Article\nPubMed/NCBI\nGoogle Scholar\n63.\nView Article\nPubMed/NCBI\nGoogle Scholar\n64.\nView Article\nPubMed/NCBI\nGoogle Scholar\n65.\nView Article\nPubMed/NCBI\nGoogle Scholar\n66.\nView Article\nPubMed/NCBI\nGoogle Scholar\n67.\nView Article\nPubMed/NCBI\nGoogle Scholar\n68.\nView Article\nPubMed/NCBI\nGoogle Scholar\n69.\nView Article\nPubMed/NCBI\nGoogle Scholar\n70.\nView Article\nPubMed/NCBI\nGoogle Scholar\n71.\nView Article\nPubMed/NCBI\nGoogle Scholar\n72.\nView Article\nPubMed/NCBI\nGoogle Scholar\n73.\nView Article\nPubMed/NCBI\nGoogle Scholar\nNishida N, Fukuda Y, Kokuryu H, Sadamoto T, Isowa G, Honda K, et al. Accumulation of allelic loss on arms of chromosomes 13q, 16q and 17p in the\nadvanced stages of human hepatocellular carcinoma. Int J Cancer. 1992;51(6):862–8. Epub 1992/07/30. pmid:1322376.\nBerry JL, Polski A, Cavenee WK, Dryja TP, Murphree AL, Gallie BL. The RB1 Story: Characterization and Cloning of the First Tumor Suppressor Gene.\nGenes (Basel). 2019;10(11). Epub 2019/11/07. pmid:31683923.\nDi Fiore R, D’Anneo A, Tesoriere G, Vento R. RB1 in cancer: different mechanisms of RB1 inactivation and alterations of pRb pathway in tumorigenesis. J\nCell Physiol. 2013;228(8):1676–87. Epub 2013/01/30. pmid:23359405.\nTakigawa Y, Brown AM. Wnt signaling in liver cancer. Curr Drug Targets. 2008;9(11):1013–24. Epub 2008/11/11. pmid:18991612.\nAoki T, Nishida N, Kudo M. Clinical Significance of the Duality of Wnt/beta-Catenin Signaling in Human Hepatocellular Carcinoma. Cancers (Basel).\n2022;14(2). Epub 2022/01/22. pmid:35053606.\nBugter JM, Fenderico N, Maurice MM. Mutations and mechanisms of WNT pathway tumour suppressors in cancer. Nat Rev Cancer. 2021;21(1):5–21.\nEpub 2020/10/25. pmid:33097916.\nMusacchio A, Salmon ED. The spindle-assembly checkpoint in space and time. Nat Rev Mol Cell Biol. 2007;8(5):379–93. Epub 2007/04/12.\npmid:17426725.\nDominguez-Brauer C, Thu KL, Mason JM, Blaser H, Bray MR, Mak TW. Targeting Mitosis in Cancer: Emerging Strategies. Mol Cell. 2015;60(4):524–36.\nEpub 2015/11/23. pmid:26590712.\nLiu X, Liao W, Yuan Q, Ou Y, Huang J. TTK activates Akt and promotes proliferation and migration of hepatocellular carcinoma cells. Oncotarget.\n2015;6(33):34309–20. Epub 2015/09/30. pmid:26418879.\nAnderhub SJ, Mak GW, Gurden MD, Faisal A, Drosopoulos K, Walsh K, et al. High Proliferation Rate and a Compromised Spindle Assembly Checkpoint\nConfers Sensitivity to the MPS1 Inhibitor BOS172722 in Triple-Negative Breast Cancers. Mol Cancer Ther. 2019;18(10):1696–707. Epub 2019/10/03.\npmid:31575759.\nChan CY, Chiu DK, Yuen VW, Law CT, Wong BP, Thu KL, et al. CFI-402257, a TTK inhibitor, effectively suppresses hepatocellular carcinoma. Proc Natl\nAcad Sci U S A. 2022;119(32):e2119514119. Epub 2022/08/02. pmid:35914158.\nLeek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package for removing batch effects and other unwanted variation in high-throughput\nexperiments. Bioinformatics. 2012;28(6):882–3. Epub 2012/01/20. pmid:22257669.\nMorris TJ, Butcher LM, Feber A, Teschendorff AE, Chakravarthy AR, Wojdacz TK, et al. ChAMP: 450k Chip Analysis Methylation Pipeline. Bioinformatics.\n2014;30(3):428–30. Epub 2013/12/18. pmid:24336642.\nSilva TC, Coetzee SG, Gull N, Yao L, Hazelett DJ, Noushmehr H, et al. ELMER v.2: an R/Bioconductor package to reconstruct gene regulatory networks\nfrom DNA methylation and transcriptome profiles. Bioinformatics. 2019;35(11):1974–7. Epub 2018/10/27. pmid:30364927.\nRu Y, Kechris KJ, Tabakoff B, Hoffman P, Radcliffe RA, Bowler R, et al. The multiMiR R package and database: integration of microRNA-target\ninteractions along with their disease and drug associations. Nucleic Acids Res. 2014;42(17):e133. Epub 2014/07/27. pmid:25063298.\nHuang JK, Jia T, Carlin DE, Ideker T. pyNBS: a Python implementation for network-based stratification of tumor mutations. Bioinformatics.\n2018;34(16):2859–61. Epub 2018/04/03. pmid:29608663.\nWu T, Hu E, Xu S, Chen M, Guo P, Dai Z, et al. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb).\n2021;2(3):100141. Epub 2021/09/25. pmid:34557778.\nHanzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics. 2013;14:7. Epub\n2013/01/18. pmid:23323831.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n15/16\n\n--- Page 16 ---\n74.\nView Article\nPubMed/NCBI\nGoogle Scholar\n75.\nView Article\nPubMed/NCBI\nGoogle Scholar\n76.\nView Article\nPubMed/NCBI\nGoogle Scholar\nKohlruss M, Reiche M, Jesinghaus M, Grosser B, Slotta-Huspenina J, Hapfelmeier A, et al. A microsatellite based multiplex PCR method for the detection\nof chromosomal instability in gastric cancer. Sci Rep. 2018;8(1):12551. Epub 2018/08/24. pmid:30135548 Ingelheim and has research funding from\nRoche, Astra Zeneca and Novartis. All remaining authors have declared no competing interests.\nFlinner N, Gretser S, Quaas A, Bankov K, Stoll A, Heckmann LE, et al. Deep learning based on hematoxylin-eosin staining outperforms\nimmunohistochemistry in predicting molecular subtypes of gastric adenocarcinoma. J Pathol. 2022;257(2):218–26. Epub 2022/02/05. pmid:35119111.\nHao Y, Hao S, Andersen-Nissen E, Mauck WM 3rd, Zheng S, Butler A, et al. Integrated analysis of multimodal single-cell data. Cell. 2021;184(13):3573–\n87 e29. Epub 2021/06/02. pmid:34062119.\n2/8/25, 4:41 PM\nMachine learning and multi-omics data reveal driver gene-based molecular subtypes in hepatocellular carcinoma for precision tr…\nhttps://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1012113&utm_source=chatgpt.com\n16/16", "input_file_path": "https://point9ml.s3.amazonaws.com/documents/Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf?response-content-type=application%2Fpdf&response-content-disposition=attachment%3B%20filename%3D%22Machine%20learning%20and%20multi-omics%20data%20reveal%20driver%20gene-based%20molecular%20subtypes%20in%20hepatocellular%20carcinoma%20for%20precision%20treatment%20_%20PLOS%20Computational%20Biology.pdf%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173749Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=683cccfdb1f09ccadfafa674c9f05ffb74fcbab71e4458da55d88bbf640bc284", "output_file_path": "https://point9ml.s3.amazonaws.com/outputs/text_text2026070617374416c893e52f89.txt?response-content-type=text%2Fplain&response-content-disposition=attachment%3B%20filename%3D%22text_text2026070617374416c893e52f89.txt%22&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAXZ5NGMFOGWOUMG7O%2F20260706%2Fap-south-1%2Fs3%2Faws4_request&X-Amz-Date=20260706T173755Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=a61f9698f9422da9c515a2f2f7e97b31b23f591567923cd0e9c62743db80e642"}}