Disentangling Protein Function via Decoupled Information Theoretic Selection of Key Tuning Residues
Saeed, H.; Yang, A.; Huang, W.
Show abstract
Rational protein engineering requires identifying residues that modulate function without disrupting functionality, a key challenge in protein engineering. Existing computational methods struggle to distinguish genuine functional sites from positions coevolving due to structural constraints, leading to high false-discovery rates. Here we present an information-theoretic decoupling framework that, without machine learning, isolates key tuning residues by computationally "denoising" sequence data, iteratively removing confounding evolutionary signals to reveal underlying functional sites. We validated this framework across 10 datasets spanning enzymes, fluorescent proteins, and antibodies. In a nanobody-antigen binding case study, our method identified > 25% (6/20) of verified binding-critical residues (p = 0.031), while the best of five benchmarked tools found zero. Performance was consistent across all datasets, with supervised variants achieving large effect sizes (Hedges g > 0.7, p < 0.01) and unsupervised variants also showing gains (g > 0.2, p < 0.05) over benchmarks. This interpretable framework provides a generalizable method to accelerate protein design, from focusing antibody maturation to optimizing biocatalysts.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- PLMFit : Benchmarking Transfer Learning with Protein Language Models for Protein Engineering 96%
- Data-efficient protein mutational effect prediction with weak supervision by molecular simulation and protein language models 95%
- EGRET: Edge Aggregated Graph Attention Networks and Transfer Learning Improve Protein-Protein Interaction Site Prediction 95%
Similar papers in this journal
- From Signal to Symphony: Exploring 2D Sequence Representations for Protein Function Prediction 96%
- From Proteins to Ligands: Decoding Deep Learning Methods for Binding Affinity Prediction 95%
- ProAffinity-GNN: A Novel Approach to Structure-based Protein-Protein Binding Affinity Prediction via a Curated Dataset and Graph Neural Networks 95%
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
- Protein Stability Prediction by Fine-tuning a Protein Language Model on a Mega-scale Dataset 95%
- Controllable Protein Design via Autoregressive Direct Coupling Analysis Conditioned on Principal Components 95%
- Towards a comprehensive view of the pocketome universe - biological implications and algorithmic challenges. 95%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.