ProteinGym V1: Benchmark for Fitness Prediction with 3 Million Mutations

ProteinGym V1 is a benchmark suite for evaluating protein fitness prediction and design models, created by OATML-Markslab and last updated on July 21, 2025. It includes nearly 3 million different mutations, covering both substitution and indel types. The dataset incorporates a wide variety of experimentally assayed proteins and clinically annotated mutations relevant to human disease.

Use Cases

• Benchmarking protein fitness prediction models based on substitution and indel mutation data
• Evaluating model performance on clinically annotated mutations relevant to human disease
• Training or testing models on a wide variety of experimentally assayed proteins

Strengths

• Includes nearly 3 million different mutations, providing a large-scale testbed
• Covers both substitution and indel mutation types
• Incorporates clinically annotated mutations relevant to human disease

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Dataset is split into four separate benchmarks; structure requires review of external documentation

Provenance

Source

OATML-Markslab

Collection Method

Likely aggregated from multiple experimental assays and clinical annotations.

Time Range

null

Freshness

Last updated 2025-07-21 02:14:14

Geography

null