sav / README.md
hazemessam's picture
Update README.md
bc26cc3 verified
metadata
license: afl-3.0
tags:
  - biology

SAV = Single Amino Acid Variant Task: Variant Effect Prediction

Note: DROP any example that has split=nan when training. The reason for leaving them is to keep this dataset identical to the original one.

This is a binary classification task predicting whether a single amino acid substitution (mutation) has:

Neutral: No significant effect on protein function/stability (benign/tolerated variant) Effect: Significant effect on protein function/stability (deleterious/pathogenic variant)

What This Dataset Is About This is a classic variant effect prediction or mutation effect prediction task, which is critical for:

Clinical genetics: Distinguishing disease-causing mutations from benign polymorphisms Protein engineering: Understanding which mutations are tolerated vs. disruptive Evolutionary biology: Identifying functionally important residues

Similar Benchmarks This task is similar to datasets like:

ClinVar: Human genetic variants (pathogenic vs. benign) ProteinGym: Variant effect prediction across many proteins COSMIC: Cancer mutation databases Tools like SIFT, PolyPhen-2, REVEL predict variant effects

How It Works For each protein, they likely have:

Wild-type sequence: ...ACDEFG... Mutation: Position 3, D→N Label: neutral or effect

The model must learn from the sequence context whether changing D to N at that position disrupts the protein. The Splits Explained

human.csv: Human protein variants only (clinically relevant) mixed.csv: Variants across multiple species (broader evolutionary context) only_savs.csv: Filtered to include only single-point mutations (excluding insertions/deletions/multiple mutations)

Neutral = variant is tolerated, doesn't break protein function Effect = variant affects function (could mean loss of function, gain of function, mislocalization, instability, etc.)