Research map / Probabilistic and scientific modeling / Molecules and proteins
Protein fitness and variant effect prediction
Predicts mutation effects using language models and autoregressive transformers.
71 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Molecules and proteins, keyword cluster “proteins · binding · sequence”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
22 papers in 2023–24 and 40 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 4.5% to 4.3%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2020 · Learning Mutational Semantics (NeurIPS 2020 · 4 citations)
- 2021 · Language models enable zero-shot prediction of the effects of mutations on protein function (NeurIPS 2021 · 699 citations)
- 2022 · Tranception: Protein Fitness Prediction with Autoregressive Transformers and Inference-time Retrieval (ICML 2022 · 125 citations)
- 2023 · ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design (NeurIPS 2023 · 30 citations)
- 2024 · MAPE-PPI: Towards Effective and Efficient Protein-Protein Interaction Prediction via Microenvironment-Aware Protein Embedding (ICLR 2024 · 10 citations)
- 2025 · Protein Language Model Fitness is a Matter of Preference (ICLR 2025 · 25 citations)
- 2026 · A New Paradigm for Genome-wide DNA Methylation Prediction Without Methylation Input (ICLR 2026 · 2 citations)
Earliest papers here
- Learning Mutational Semantics (NeurIPS 2020 · 4 citations)
- Fourier-transform-based attribution priors improve the interpretability and stability of deep learning models for genomics (NeurIPS 2020)
- Language models enable zero-shot prediction of the effects of mutations on protein function (NeurIPS 2021 · 699 citations)
What has been tried
42 papers here report 445 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 8%, hurt 19%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add masked codebook modeling (MCM) | SHS27k, SHS148k, and STRING (DFS/BFS; average) · micro-F1 average | 75.26 -> 76.89 | Helped | MAPE-PPI: Towards Effective and Efficient Protein-Protein Interaction Prediction via Microenvironment-Aware Protein Embedding (ICLR 2024 · 10 citations) |
| replace the MLP decoder with a Bi-RNN decoder | Time split, enzyme-to-reaction · Top1 | 0.3246 -> 0.3911 | Helped | ReactZyme: A Benchmark for Enzyme-Reaction Prediction (NeurIPS 2024 · 7 citations) |
| add PIE | SAbDab CDR-H1 · AAR | 60.27±6.69% -> 64.34±3.37% | Helped | Cross-Gate MLP with Protein Complex Invariant Embedding Is a One-Shot Antibody Designer (AAAI 2024 · 6 citations) |
| add microenvironment-aware prompts | SKEMPI v2.0 · Per-Structure Pearson | 0.4114 -> 0.4712 | Helped | Learning to Predict Mutational Effects of Protein-Protein Interactions by Microenvironment-aware Hierarchical Prompt Learning (ICML 2024 · 4 citations) |
| Use von Mises angle distributions instead of direct angle predictions | Peptide Binder Design · RMSD | 3.10 -> 2.68 | Helped | Hotspot-Driven Peptide Design via Multi-Fragment Autoregressive Extension (ICLR 2025 · 3 citations) |
| add gap state to SiteRM | ProteinGym DMS substitutions · Spearman | 0.412 -> 0.423 | Helped | Ultrafast classical phylogenetic method beats large protein language models on variant effect prediction (NeurIPS 2024 · 3 citations) |
| Use PG-AbD with FiLM rather than without FiLM | RabDab CDR-H3 · AAR | 35.17 -> 58.40 | Helped | Synergy of GFlowNet and Protein Language Model Makes a Diverse Antibody Designer (AAAI 2025 · 2 citations) |
| Add fine-tuning | ProteinGym single-mutant, 128-shot · Spearman correlation | .488 -> .552 | Helped | Metalic: Meta-Learning In-Context with Protein Language Models (ICLR 2025 · 2 citations) |
| remove the binary interaction layer | RAbD CDR-H3 · UA-RMSD | 1.5852 -> 1.6695 | Hurt | GeoAB: Towards Realistic Antibody Design and Reliable Affinity Maturation (ICML 2024 · 2 citations) |
| Use three alternating iterations rather than one iteration with multiple simultaneous mutations | CDR-H3 · IMP | 73.3% -> 93.3% | Helped | AffinityFlow: Guided Flows for Antibody Affinity Maturation (ICML 2025 · 1 citation) |
| add affinity guidance | CrossDocked · Vina | -7.375±0.10 -> -8.236±0.16 | Helped | Generalized Protein Pocket Generation with Prior-Informed Flow Matching (NeurIPS 2024 · 1 citation) |
| Increase δ | AMP and CRE · PI | Helped | GFlowNet Assisted Biological Sequence Editing (NeurIPS 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 25 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- SabDab · 2 of 25 papers
- PepBDB · 2 of 25 papers
- UniRef50 · 2 of 25 papers
Baselines
- ESM-2 · 4 of 25 papers
- ESM-1v · 3 of 25 papers
- refine-GNN · 2 of 25 papers
- RFDiffusion · 2 of 25 papers
- ProteinGenerator · 2 of 25 papers
- PepFlow · 2 of 25 papers
- PepGLAD · 2 of 25 papers
- ProGen2 · 2 of 25 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 62 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 6 papers
- Harvard University · 5 papers
- Westlake University · 4 papers
- University of Oxford · 4 papers
- Alibaba Group · 3 papers
- Tsinghua University · 3 papers
- Genentech · 3 papers
- University of Science and Technology of China · 2 papers
- University of Notre Dame · 2 papers
- University of Illinois, Urbana Champaign · 2 papers
Use it
Download BibTeX of 34 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Generative protein structure design (142 papers)
- Protein interface and docking prediction (72 papers)
- Protein representation learning (40 papers)
- Biological sequence transformers (34 papers)
Elsewhere in the topic
- Prediction-powered and conformal inference (129 papers)
- Statistical analysis of ML practice (122 papers)
- Neuro-symbolic probabilistic layers (120 papers)
- Structure-based 3D molecule generation (107 papers)
- Sparse graphical and tensor models (97 papers)
- Flow matching generative modeling (97 papers)
- Machine learning for physical simulation (96 papers)
- Multimodal trajectory prediction (95 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).