Research map / Probabilistic and scientific modeling / Molecules and proteins
Language models enable zero-shot prediction of the effects of mutations on protein function
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| ensemble five ESM-1v models (ablation) | ESM-1v · 41 DMS datasets (full) | average |Spearman rho|: 0.484 -> 0.509 | Helped |
| spiked fine-tuning on MSA regularized with pre-training sequences (ablation) | ESM-1v · 41 DMS datasets (full) | average |Spearman rho|: 0.510 -> 0.537 | Helped |
| naive fine-tuning on the MSA (ablation) | ESM-1v · DMS datasets | average |Spearman rho| | Hurt |
| pretrain on UniRef clustered at 90% instead of 50% identity (ablation) | ESM-1v · DMS validation set | average |Spearman rho| | Helped |
| pretrain on unclustered Uniref100 (ablation) | ESM-1v · DMS validation set | average |Spearman rho| | Hurt |
| masked marginal scoring instead of other scoring methods (ablation) | ESM-1v · DMS validation set | average |Spearman rho| | Helped |
| increase parameter count (ablation) | protein language models · DMS datasets | average |Spearman rho| | Helped |
| sample MSA sequences by sequence weights instead of diversity-minimizing selection (ablation) | MSA Transformer · DMS datasets | average |Spearman rho| | Helped |
| zero-shot ESM-1v instead of family-specific DeepSequence | variant effect prediction · 41 DMS datasets (full) | average |Spearman rho|: 0.520 -> 0.509 | No clear effect |
Closest papers by meaning
- DePLM: Denoising Protein Language Models for Property Optimization (NeurIPS 2024 · similarity 0.89)
- MutaPLM: Protein Language Modeling for Mutation Explanation and Engineering (NeurIPS 2024 · similarity 0.89)
- Understanding protein function with a multimodal retrieval-augmented foundation model (NeurIPS 2025 · similarity 0.89)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.88)
- From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models (ICML 2025 · similarity 0.88)
- Steering Protein Language Models (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.