Biophysics Enhances Deep Learning Protein Variant Prediction
A team led by Jianhan Chen (Barethiya et al.) on March 25, 2026 reports that integrating biophysics-derived energetics into convolutional and graph neural networks improves extrapolation in protein sequence-to-function prediction. They show stability-derived energetic features boost positional and mutational extrapolation on deep mutational scanning datasets and that adding protein language model evolutionary embeddings further improves accuracy. The approach yields more reliable variant effect predictors for protein engineering and genetic variation prioritization.
Key Points
- 1Incorporates biophysics-derived mutation energetics into convolutional and graph neural networks
- 2Improves positional and mutational extrapolation on deep mutational scanning benchmarks
- 3Enables more accurate variant effect predictors for protein engineering and disease prioritization
Scoring Rationale
High methodological novelty, broad applicability, and reproducibility raise impact; limited to evaluated DMS datasets and protein types.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
