Biology+AI Daily@BiologyAIDaily
Leveraging Unified Sequence-Structure Representations for Enhanced Protein Stability Prediction
1. A new deep learning framework named ProStab-Former has been introduced to predict protein thermal stability with high accuracy. This model integrates sequence and structure information in a unified representation space, overcoming limitations of previous multi-modal approaches that struggled with indirect information fusion and incomplete capture of sequence-structure interactions.
2. ProStab-Former leverages a pre-trained, multi-modal protein foundation encoder for robust feature extraction at the residue level. It includes Stability-Aware Attention Layers (SAAL) with a structural prior bias and mutation-aware gating, enabling precise capture of local and global stability perturbations induced by mutations.
3. The model incorporates an Epistatic Interaction Module to explicitly model non-additive effects in multi-point mutations. This allows ProStab-Former to efficiently predict the stability changes for numerous single-point mutations in a single pass, significantly accelerating mutation landscape analysis.
4. Experimental evaluations show that ProStab-Former achieves superior performance, with a median Spearman correlation coefficient of 0.84 on the Megascale Test set, surpassing state-of-the-art models like SPURS. It also demonstrates strong generalization across diverse tasks, including melting temperature prediction and pathogenic mutation classification.
5. ProStab-Former's design prioritizes computational efficiency, enabling rapid inference for comprehensive mutation scanning. This makes it highly practical for high-throughput protein engineering and variant effect analysis, positioning it as a valuable tool for both research and industrial applications.
📜Paper: biorxiv.org/content/10.648…
#ProteinStability #DeepLearning #ProteinEngineering #Bioinformatics
