Leaderboard Dataset Overview
We are excited to present a comprehensive biomedical dataset curated for advanced research and predictive modeling. This collection provides rich, multi-dimensional data across diverse cancer types, offering a solid foundation for developing robust machine learning models.
📊 Key Statistics
- Patients: train: 81 individuals, test: 34 individuals
- Total Records: train: 85,962 data items, test: 36,243 data items
- Average Data per Patient: ~1,100 entries
🎯 Cancer Type Coverage
The dataset encompasses 10 distinct cancer types, providing broad representation across oncology domains:
- Breast Cancer
- Clear Cell Renal Cell Carcinoma
- Colorectal Adenocarcinoma
- Glioblastoma (GBM)
- Head & Neck SCC
- Leukemia
- NSCLC (Non-Small Cell Lung Cancer)
- Ovarian Cancer
- PDAC (Pancreatic Ductal Adenocarcinoma)
- Skin Cutaneous Melanoma
Data Feature Description
Training Data Notes
Unlike the NeoRanking dataset which includes 31 features, our training data only utilizes the following 8 features, with explanations provided below.
Features
- seq_len: Peptide Length
- mutant_rank: MixMHCpred Rank
- mutant_other_significant_alleles: Number of Binding Alleles
- TAP_score: NetTAP Score
- mutant_rank_netMHCpan: NetMHCpan Rank
- mut_Rank_Stab: NetStab Rank
- mutant_rank_PRIME: PRIME Rank
- mut_netchop_score_ct: NetChop CT Score
Tools and Versions
The features were calculated using the following tools and versions:
- MixMHCpred-2.1
- netchop-3.1
- PRIME-1.0
- netMHCstabpan-1.0
- netMHCpan-4.1
- netCTLpan-1.1
Additional Notes
- The training data is labeled with positive and negative samples.
- You may combine this data with your own datasets or use it directly for training.
- For the ranking task, there are no specific requirements on the training data — you only need to provide predictions for the given test set.
🚀 Getting Started
The dataset is now available for download. We encourage researchers to explore the rich phenotypic and genotypic information contained within this collection. The balanced nature and comprehensive coverage make it ideal for developing predictive models that can generalize across multiple cancer types.
We look forward to seeing your innovative approaches and wish you the best of luck on the leaderboard! Your contributions will help advance our understanding of cancer immunology and improve patient outcomes.