SunnyLin/clinvar-summary

Dataset Card for clinvar_variant_summary_parquet

This dataset contains the processed and compressed Parquet version of the variant_summary.txt file from the ClinVar database, which archives relationships between human variations and health.

Dataset Details

Dataset Description

The original ClinVar database is a public archive of reports on the relationships between human variations and health, maintained by the National Institutes of Health (NIH).

This dataset provides the full contents of the official variant_summary.txt file, which is the core table summarizing all unique variants in ClinVar. It is converted from the original GZIP-compressed, tab-delimited text format into the Parquet format. Parquet is a column-oriented, compressed file format, offering improved storage efficiency and faster analytical querying compared to the original text file.

Key information included for each variant: Variant ID, Gene Symbol, Name (HGVS notation), Clinical Significance (Pathogenic, Benign, VUS, etc.), Review Status, and submission counts.

Dataset Sources

• Source Data (Typical FTP Location): ftp://ftp.ncbi.nlm.nih.gov/pub/clinvar/tab_delimited/variant_summary.txt.gz

Uses

Direct Use

This dataset is suitable for:

• Bioinformatics Research: Studying the prevalence and characteristics of variants categorized by clinical significance.
• Data Analysis and Visualization: Creating summary statistics and visualizations of the ClinVar data using efficient data analysis tools (e.g., Pandas, Spark).
• Machine Learning: Training models for variant pathogenicity prediction, provided the user carefully handles the data limitations (see Out-of-Scope Use).

Out-of-Scope Use

• Direct Diagnostic Use or Medical Decision-Making: As stated by the NCBI, ClinVar data is not intended for direct diagnostic use or medical decision-making without review by a genetics professional. The dataset should not be used to interpret a patient's genetic status in a clinical setting.
• Clinical Advice Generation: Models trained on this data should not be used to generate clinical advice without validation and oversight from qualified professionals.
• Inferring Patient Identities: While the variant_summary.txt file is aggregated, the source data contains information submitted by various clinics and labs. Any attempt to link this aggregated data back to individual patients is strictly out-of-scope and unethical.

Dataset Structure

The dataset is provided as a single split named train.

The data fields directly correspond to the columns in the original variant_summary.txt file.

Column Name	Data Type	Description
VariationID	int64	ClinVar ID for the variant.
ClinicalSignificance	string	The consensus clinical interpretation (e.g., Pathogenic, Benign, VUS - Variant of Uncertain Significance).
ReviewStatus	string	The level of evidence/consensus (e.g., criteria provided, multiple submissions, expert panel).
GeneSymbol	string	The associated gene symbol.
Name	string	The variant name, typically in HGVS notation.
NumberSubmitters	int64	Count of unique organizations that have submitted an interpretation.
NumberSubmissions	int64	Total number of interpretation submissions.
DateLastEvaluated	string	Date the variant was last assessed.
...	...	All other columns from the original file are retained.

Dataset Creation

Curation Rationale

The motivation for creating this Parquet-formatted Hugging Face dataset is to provide a highly efficient, compressed, and easily loadable version of the vast ClinVar summary data. This conversion greatly improves data loading times and reduces memory usage, making the data more accessible for analytical and machine learning workloads, especially for users utilizing cloud environments or resource-constrained systems.

Source Data

The source data is the complete variant_summary.txt file from the official ClinVar FTP repository. This file is an aggregation and summary of individual submissions from clinical and research laboratories across the globe.

Data Collection and Processing

1. Download: The variant_summary.txt.gz file was downloaded from the NCBI ClinVar FTP server.
2. Conversion: The file was decompressed, read as a tab-separated value (TSV) file, and then converted directly to the Parquet format. Data types were inferred during this conversion (e.g., numerical IDs as integers, dates and names as strings).
3. Tooling: Standard Python data processing libraries (e.g., pandas, pyarrow) were used for the conversion.
4. No Filtering or Normalization: No rows or columns were dropped, and no significant value-level normalization was performed to ensure maximum fidelity to the original source file.

Who are the source data producers?

The data is produced by thousands of clinical and research laboratories, professional organizations, and individuals worldwide who submit interpretations of variants to ClinVar. The final, curated summary file is assembled and maintained by the National Center for Biotechnology Information (NCBI) at the NIH.

Personal and Sensitive Information

The variant_summary.txt file does not contain personally identifiable information (PII) of patients. It consists of aggregated, de-identified genetic and molecular data, along with metadata about the submissions.

Bias, Risks, and Limitations

Bias:

• Geographic Bias: Submissions may be biased towards populations primarily studied by Western (US/European) clinical labs and research institutions, potentially underrepresenting variants common in other global populations.
• Pathogenicity Skew: There is a known bias in ClinVar towards submitting variants that are known or suspected to be pathogenic, leading to an imbalance in the data relative to benign variation.
• Reporting Bias: Variant interpretations are based on the knowledge available at the time of submission and the specific testing methodologies used by the submitting laboratories.

Limitations:

• Frequency of Updates: ClinVar is updated frequently. This dataset is a static snapshot and will become outdated over time. Users should check the source date for relevance.
• Discrepancies: The database contains conflicting interpretations for some variants. The ClinicalSignificance field often reflects a consensus, but users should consult the ReviewStatus and submission counts.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset.

• Consult Source Documentation: Always refer to the official NCBI ClinVar documentation for the most accurate and up-to-date definitions of columns and data usage policies.
• Use for Research ONLY: Do not use this dataset for clinical diagnosis.
• Address Class Imbalance: When training machine learning models, take steps to mitigate the severe class imbalance in the ClinicalSignificance field (e.g., over-sampling of rare classes, appropriate loss functions).

Citation

If you use the data in a publication, please cite the official ClinVar resource.

More Information

• NCBI ClinVar Home Page