FLIP
A supervised protein sequence-to-fitness benchmark that turns three experimental landscapes into 15 biologically motivated dataset splits for testing generalization in protein engineering.
Scientific task · Property and function prediction
Protein fitness prediction
Predict a quantitative or ranked protein fitness measurement from sequence.
蛋白质适应度预测
Definition and search aliases
- Permanent ID
protein-fitness-prediction- Aliases
- fitness landscape prediction, sequence-to-fitness
- Deprecated aliases
- None
- Hierarchy
- Leaf task under Protein property and function prediction
Coverage
2 benchmark families cover this task
FLIP AAV
Seven supervised splits over sampled and machine-designed AAV2 VP-1 capsid variants, measuring generalization across mutation depth, fitness, and sampled-versus-designed pools.
FLIP GB1
Five supervised splits over a downsampled, highly epistatic four-site GB1 immunoglobulin-binding landscape, designed to test mutation-depth and low-to-high-fitness generalization.
ProteinGym
Versioned deep-mutational-scanning and clinical-variant benchmarks for protein fitness prediction and design in zero-shot and supervised regimes.
ProteinGym DMS Indels
ProteinGym track for predicting experimental fitness measurements of insertion and deletion mutants across deep-mutational-scanning assays.
ProteinGym DMS Substitutions
ProteinGym track for predicting experimental fitness measurements of substitution mutants across deep-mutational-scanning assays.
Evidence-backed count claims
Each row keeps its original unit and basis. Rows with different units or overlapping mappings are never added.
| Benchmark | Mapped task | Coverage | Count | Version | Evidence |
|---|---|---|---|---|---|
| FLIP root: flip | Protein fitness prediction official-taxonomy · high | explicitly-in-scope | 15 tasks Dataset-and-split benchmark tasks. | original-2021 | flip-evidence-paper-definition |
| FLIP AAV root: flip | Protein fitness prediction official-track · high | explicitly-in-scope | 284009 examples Distinct sequence-fitness examples across the sampled and designed AAV pools. | original-2021 | flip-aav-evidence-paper |
| FLIP GB1 root: flip | Protein fitness prediction official-track · high | explicitly-in-scope | 8733 examples Downsampled sequence-fitness examples retained for FLIP. | original-2021 | flip-gb1-evidence-paper |
| ProteinGym root: proteingym | Protein fitness prediction official-taxonomy · high | explicitly-in-scope | Not reported ProteinGym DMS assays. | 1.3 | proteingym-evidence-taxonomy |
| ProteinGym DMS Indels root: proteingym | Protein fitness prediction official-track · high | explicitly-in-scope | 66 assays DMS assays. | 1.3 | proteingym-dms-indel-evidence-v13 |
| ProteinGym DMS Substitutions root: proteingym | Protein fitness prediction official-track · high | explicitly-in-scope | 217 assays DMS assays. | 1.3 | proteingym-dms-sub-evidence-v13 |
Official evaluations connected to these benchmarks
Runs are included only for benchmark records mapped here (and formal child tracks when a mapped suite is the root). A task mapping does not imply that every run isolates this task.
| Work | Provider / class | Related runs |
|---|---|---|
| FLIP: Benchmark tasks in fitness landscape inference for proteins | Technical University of Munich, Microsoft Research New England, California Institute of Technology, University of California Berkeley, Massachusetts Institute of Technology, Salesforce Research benchmark_creator | flip-aav-des-mutflip-aav-low-vs-highflip-aav-mut-desflip-aav-one-vs-restflip-aav-sampledflip-aav-seven-vs-restflip-aav-two-vs-restflip-gb1-low-vs-highflip-gb1-one-vs-restflip-gb1-sampledflip-gb1-three-vs-restflip-gb1-two-vs-restflip-meltome-humanflip-meltome-human-cellflip-meltome-mixed |
| ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design | University of Oxford, Harvard Medical School, Seismic Therapeutic, Harvard University, Centre for Genomic Regulation, Universitat Pompeu Fabra, Broad Institute benchmark_creator | proteingym-v10-dms-substitutions-zero-shot |