Enhancing protein structure prediction accuracy by prioritizing important residues using protein language models

Accurate prediction of protein tertiary structures from amino acid sequences remains a fundamental challenge in computational biology. Although AlphaFold2 represents a major advance, systematic discrepancies persist between its predictions and experimentally determined structures. Given that individual residues contribute differentially to protein function, we hypothesize that incorporating residue-specific importance metrics improves prediction accuracy. Here, we develop i-Fold (importance Fold), an enhanced neural architecture that builds upon the AlphaFold2 framework by integrating protein language model-derived residue importance scores as dynamic positional weights during training. Evaluation on a benchmark test set of 3599 protein structures reveals that the prediction error of i-Fold is reduced by 0.3 Å on average, improving the prediction success rate by 7.6%, and consistent results are obtained on an independent test set of 167 recently released protein structures. Notably, i-Fold demonstrates particular improvements for targets that are typically challenging for AlphaFold2, including ribosomal proteins, membrane proteins, and orphan proteins. Our findings indicate that explicit integration of evolutionary residue importance can advance the state-of-the-art in protein structure prediction, producing more accurate and generalizable models without substantially increasing computational cost. The i-Fold architecture integrates language modelderived residue importance scores to guide AlphaFold2, significantly improving structure prediction accuracy and resolving topological bottlenecks in flexible protein regions.