Upload folder using huggingface_hub

.gitattributes

@@ -57,3 +57,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+2023/meta.jsonl filter=lfs diff=lfs merge=lfs -text
+2023/pmid_to_citation.json filter=lfs diff=lfs merge=lfs -text
+2023/pmid_to_meta_offset.json filter=lfs diff=lfs merge=lfs -text
+2025/meta_info_2025_0327_part1.csv filter=lfs diff=lfs merge=lfs -text
+2025/meta_info_2025_0327_part2.csv filter=lfs diff=lfs merge=lfs -text

2023/meta.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d944bf7e0d0a9c490e4684492efd20ec6c2a7641e2cf6d8166b1f6fbf047c5f1
+size 7925257329

2023/pmid_to_citation.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44b21bb4458488a0393cf2d4c55b921aedd68b19f27297bccd13b99371a7f071
+size 404894565

2023/pmid_to_meta_offset.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d255923d407a0f10172fda82396fa6983a2604bcfeec681813c00225ed2d0f8c
+size 834226190

2025/merge_large_data.py

@@ -0,0 +1,13 @@
+import pandas as pd
+
+# 读取拆分后的 CSV 文件
+df1 = pd.read_csv('meta_info_2025_0327_part1.csv')
+df2 = pd.read_csv('meta_info_2025_0327_part2.csv')
+
+# 合并数据框
+df_merged = pd.concat([df1, df2], ignore_index=True)
+
+# 保存合并后的 CSV 文件
+df_merged.to_csv('meta_info_2025_0327.csv', index=False)
+
+print("CSV 文件已成功合并！")

2025/meta_info_2025_0327_part1.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a881cb2c324f11c5ecd3c4789b96f49bbfb10c6484b6b59c7349c1abd75150e6
+size 24783788188

2025/meta_info_2025_0327_part2.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20ca28405daf4d59dfef1b361166d4675d323898cfbf78583675767510eacc35
+size 25430171535

2025/split_large_data.py

@@ -0,0 +1,17 @@
+import pandas as pd
+
+# 读取原始 CSV 文件
+df = pd.read_csv('meta_info_2025_0327.csv')
+
+# 计算拆分点，假设将文件拆分为两半
+split_index = len(df) // 2
+
+# 拆分数据帧
+df1 = df.iloc[:split_index]  # 前半部分
+df2 = df.iloc[split_index:]  # 后半部分
+
+# 保存拆分后的 CSV 文件
+df1.to_csv('meta_info_2025_0327_part1.csv', index=False)
+df2.to_csv('meta_info_2025_0327_part2.csv', index=False)
+
+print("CSV 文件拆分完成！")

download.py

@@ -0,0 +1,105 @@
+import json
+import os
+import ftplib
+import hashlib
+import logging
+from time import sleep
+
+# 配置参数
+ftp_url = "download.nmdc.cn"
+ftp_dir = "/pubmed"
+local_dir = "./pubmed_data"
+file_patterns = (".xml.gz", ".md5")  # 需要下载的文件类型
+max_retries = 3  # 单个文件下载失败重试次数
+retry_delay = 10  # 重试间隔(秒)
+
+# 初始化日志
+logging.basicConfig(
+    filename='pubmed_download.log',
+    level=logging.INFO,
+    format='%(asctime)s - %(levelname)s - %(message)s',
+    encoding='utf-8'
+)
+
+def verify_md5(file_path, expected_md5):
+    """验证文件的MD5校验码"""
+    hash_md5 = hashlib.md5()
+    with open(file_path, "rb") as f:
+        for chunk in iter(lambda: f.read(4096), b""):
+            hash_md5.update(chunk)
+    return hash_md5.hexdigest() == expected_md5
+
+def download_file(ftp, filename, local_path):
+    """带重试机制的文件下载"""
+    for attempt in range(max_retries):
+        try:
+            with open(local_path, 'wb') as f:
+                ftp.retrbinary(f"RETR {filename}", f.write)
+            return True
+        except Exception as e:
+            logging.error(f"Attempt {attempt+1} failed for {filename}: {str(e)}")
+            if attempt < max_retries - 1:
+                sleep(retry_delay)
+    os.remove(local_path)
+    return False
+
+def main():
+    # 创建本地目录
+    os.makedirs(local_dir, exist_ok=True)
+    error_files = []
+    try:
+        # 连接FTP
+        ftp = ftplib.FTP(ftp_url, timeout=30)
+        ftp.login()  # 匿名登录
+        ftp.cwd(ftp_dir)
+        
+        # 获取文件列表
+        files = [f for f in ftp.nlst() if f.endswith(file_patterns) and f.startswith('pubmed25')]
+        logging.info(f"Found {len(files)} files to process")
+
+        # 下载主循环
+        for idx, filename in enumerate(files, 1):
+            try:
+                local_path = os.path.join(local_dir, filename)
+                
+                # 跳过已存在的完整文件
+                if os.path.exists(local_path):
+                    logging.info(f"Skipping existing file: {filename}")
+                    continue
+
+                # 下载文件
+                logging.info(f"Downloading ({idx}/{len(files)}) {filename}")
+                success = download_file(ftp, filename, local_path)
+                
+                # MD5校验
+                if success and filename.endswith(".md5"):
+                    continue  # 不需要校验MD5文件本身
+                else:
+                    logging.info(f"Downloading MD5 for {filename}")
+                    md5_file = filename + ".md5"
+                    md5_local_path = os.path.join(local_dir, md5_file)
+                    success2 = download_file(ftp, md5_file, md5_local_path)
+                    
+                if success and success2 and filename.endswith(".xml.gz"):
+                    md5_file = filename + ".md5"
+                    if md5_file in files:
+                        md5_path = os.path.join(local_dir, md5_file)
+                        with open(md5_path) as f:
+                            expected_md5 = f.read().split()[1].replace('\n', '')
+                        if not verify_md5(local_path, expected_md5):
+                            logging.error(f"MD5 mismatch for {filename}")
+                            os.remove(local_path)  # 删除损坏文件
+            except Exception as e:
+                logging.error(f"{filename}:Fatal error: {str(e)}")
+                error_files.append(filename)
+
+    except Exception as e:
+        logging.error(f"Fatal error: {str(e)}")
+    finally:
+        if 'ftp' in locals():
+            ftp.quit()
+        with open('./error_file.json', 'w') as f:
+            json.dump(error_files, f)
+
+if __name__ == "__main__":
+    main()

process_metadata.py

@@ -0,0 +1,171 @@
+import os
+import gzip
+import xml.etree.ElementTree as ET
+import pandas as pd
+import tqdm
+import glob
+
+import xml.etree.ElementTree as ET
+
+def extract_meta_info(xml_content):
+    root = ET.fromstring(xml_content)
+    meta_info_list = []  # List to hold metadata of each article
+
+    # Loop over each article in the XML (assuming it's in a root <PubmedArticle> list)
+    articles = root.findall(".//PubmedArticle")  # Or adjust the XPath based on your XML structure
+    
+    for article in articles:
+        meta_info = {}
+        
+        # Extract PMID
+        pmid = article.find(".//PMID")
+        meta_info['PMID'] = pmid.text if pmid is not None else None
+        
+        # Extract DateCompleted
+        date_completed = article.find(".//DateCompleted")
+        if date_completed is not None:
+            year = date_completed.find(".//Year")
+            month = date_completed.find(".//Month")
+            day = date_completed.find(".//Day")
+            meta_info['DateCompleted'] = f"{year.text}-{month.text}-{day.text}" if year is not None and month is not None and day is not None else None
+        
+        # Extract DateRevised
+        date_revised = article.find(".//DateRevised")
+        if date_revised is not None:
+            year = date_revised.find(".//Year")
+            month = date_revised.find(".//Month")
+            day = date_revised.find(".//Day")
+            meta_info['DateRevised'] = f"{year.text}-{month.text}-{day.text}" if year is not None and month is not None and day is not None else None
+        
+        # Extract ISSN
+        issn = article.find(".//ISSN")
+        meta_info['ISSN'] = issn.text if issn is not None else None
+        
+        # Extract Journal Title
+        journal_title = article.find(".//Journal/Title")
+        meta_info['JournalTitle'] = journal_title.text if journal_title is not None else None
+        
+        # Extract Article Title
+        article_title = article.find(".//ArticleTitle")
+        meta_info['ArticleTitle'] = article_title.text if article_title is not None else None
+        
+        # Extract Authors
+        authors = article.findall(".//AuthorList/Author")
+        author_names = []
+        for author in authors:
+            last_name = author.find(".//LastName")
+            fore_name = author.find(".//ForeName")
+            if last_name is not None and fore_name is not None:
+                author_names.append(f"{last_name.text} {fore_name.text}")
+        meta_info['Authors'] = ', '.join(author_names) if author_names else None
+        
+        # Extract Language
+        language = article.find(".//Language")
+        meta_info['Language'] = language.text if language is not None else None
+        
+        # Extract Grants
+        grants = article.findall(".//GrantList/Grant")
+        grant_info = []
+        for grant in grants:
+            grant_id = grant.find(".//GrantID")
+            agency = grant.find(".//Agency")
+            country = grant.find(".//Country")
+            if grant_id is not None and agency is not None and country is not None:
+                grant_info.append(f"{grant_id.text} ({agency.text}, {country.text})")
+        meta_info['Grants'] = '; '.join(grant_info) if grant_info else None
+        
+        # Extract Publication Types
+        publication_types = article.findall(".//PublicationTypeList/PublicationType")
+        pub_types = []
+        for pub_type in publication_types:
+            pub_types.append(pub_type.text)
+        meta_info['PublicationTypes'] = ', '.join(pub_types) if pub_types else None
+        
+        # Extract Chemicals
+        chemicals = article.findall(".//ChemicalList/Chemical")
+        chemical_info = []
+        for chemical in chemicals:
+            substance_name = chemical.find(".//NameOfSubstance")
+            if substance_name is not None:
+                chemical_info.append(substance_name.text)
+        meta_info['Chemicals'] = ', '.join(chemical_info) if chemical_info else None
+        
+        # Extract CitationSubset
+        citation_subset = article.find(".//CitationSubset")
+        meta_info['CitationSubset'] = citation_subset.text if citation_subset is not None else None
+        
+        # Extract Article IDs (DOI, etc.)
+        article_ids = article.findall(".//ArticleIdList/ArticleId")
+        article_id_info = []
+        for article_id in article_ids:
+            article_id_info.append(article_id.text)
+        meta_info['ArticleIds'] = ', '.join(filter(None, article_id_info)) if article_id_info else None
+        
+        # Extract Abstract
+        abstract = article.find(".//Abstract/AbstractText")
+        meta_info['Abstract'] = abstract.text if abstract is not None else None
+        
+        # Extract Mesh Terms
+        mesh_terms = article.findall(".//MeshHeadingList/MeshHeading")
+        mesh_terms_info = []
+        for mesh_term in mesh_terms:
+            descriptor_name = mesh_term.find(".//DescriptorName")
+            if descriptor_name is not None:
+                mesh_terms_info.append(descriptor_name.text)
+        meta_info['MeshTerms'] = ', '.join(filter(None, mesh_terms_info)) if mesh_terms_info else None
+        
+        # Extract Keywords
+        keywords = article.findall(".//KeywordList/Keyword")
+        keyword_info = []
+        for keyword in keywords:
+            keyword_info.append(keyword.text)
+        meta_info['Keywords'] = ', '.join(filter(None, keyword_info)) if keyword_info else None
+        
+        # Append the metadata for this article to the list
+        meta_info_list.append(meta_info)
+    
+    return meta_info_list
+
+def extract(input_dir, output_csv):
+    # Create a temporary directory to store individual CSVs
+    temp_dir = os.path.join(os.path.dirname(output_csv), 'temp')
+    os.makedirs(temp_dir, exist_ok=True)
+    
+    # Iterate over all .gz files in the directory
+    for filename in tqdm.tqdm(os.listdir(input_dir)):
+        if filename.endswith('.xml.gz'):
+            file_path = os.path.join(input_dir, filename)
+            
+            # Decompress and read the XML content
+            with gzip.open(file_path, 'rb') as f:
+                xml_content = f.read()
+            
+            # Extract meta information
+            meta_info_list = extract_meta_info(xml_content)
+
+            # Save meta information to a temporary CSV file
+            temp_csv_path = os.path.join(temp_dir, f"{os.path.splitext(filename)[0]}.csv")
+
+            # Create a DataFrame from the list of dictionaries (each dict represents an article's metadata)
+            df = pd.DataFrame(meta_info_list)
+
+            # Save the DataFrame to a CSV file
+            df.to_csv(temp_csv_path, index=False)
+    
+    # Combine all temporary CSVs into a single large CSV file
+    all_csv_files = glob.glob(os.path.join(temp_dir, '*.csv'))
+    combined_df = pd.concat((pd.read_csv(f) for f in all_csv_files), ignore_index=True)
+    combined_df.to_csv(output_csv, index=False)
+    
+    # Optionally, delete the temporary files
+    # for f in all_csv_files:
+    #     os.remove(f)
+    # os.rmdir(temp_dir)
+    
+    print(f"Meta information extracted and saved to {output_csv}")
+
+if __name__ == "__main__":
+    # Define the input directory
+    input_dir = './pubmed_data'  # Replace with actual path
+    output_csv = './2025/meta_info_2025_0327.csv'  # Output CSV file path
+    extract(input_dir=input_dir, output_csv=output_csv)

readme.md

@@ -0,0 +1,15 @@
+# 1. origin source
+https://ftp.ncbi.nlm.nih.gov/pubmed/baseline/
+
+# 2. 2023 data sources
+https://github.com/jacobvsdanniel/pubmedkb_web 
+*We processed all 35M PubMed citations dumped on 2023/02/17.
+https://drive.google.com/file/d/17AUaMe0w3xJq3rr0Njs1tok2y3RS5tQ2/view
+
+# 3. 2025 data sources
+
+https://www.nmdc.cn/datadownload 
+
+[download.py](download.py) ： need monthly update  
+
+[process_meatadata.py](process_meatadata.py) : need process xml to csv