FLIP2: Second-Generation Protein Fitness Landscape Benchmark

Released in February 2026, FLIP2 is a second-generation benchmark for protein fitness landscape inference. It expands upon the original FLIP benchmark by covering seven new datasets, including enzymes, and was created by a group spanning NVIDIA, Microsoft, Caltech, Profluent, and Duke.

Use Cases

• Benchmarking protein fitness prediction models based on the expanded scope to seven datasets.
• Training models for enzyme activity prediction based on the description mentioning enzyme datasets.
• Comparing model performance on thermostability, binding, and viability tasks based on the predecessor benchmark's focus.

Strengths

• Expands scope to seven new datasets, broadening the benchmark's coverage.
• Created by a multi-institutional collaboration including NVIDIA, Microsoft, Caltech, Profluent, and Duke.
• Released as a second-generation update, superseding a previous NeurIPS 2021 benchmark.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

LiteFold on Hugging Face

Freshness

Last updated 2026-05-27 12:47:12; freshness should be verified.