DSSA-PPI: enhancing binding affinity change prediction upon protein mutations using disentangled structure-sequence...

Area of Science:

• Computational Biology
• Biochemistry
• Genomics

Background:

• Accurate prediction of mutation effects on protein-protein interactions (PPIs) is vital for disease mechanism research and targeted therapy development.
• Existing methods often struggle to precisely capture the complex interplay of sequence and structural data in predicting these effects.

Purpose of the Study:

• To introduce DSSA-PPI, a novel hybrid deep learning framework designed to enhance the prediction of mutation-induced binding affinity changes (ΔΔG) in PPIs.
• To integrate structural and sequence information effectively using a disentangled attention mechanism for improved prediction accuracy.

Main Methods:

• Developed DSSA-PPI, a hybrid deep learning framework combining a geometric equivariant graph neural network (PPIFormer) and a protein language model (ESM-2).
• Employed a novel representation learning strategy to integrate sequence- and structure-specific contributions.
• Validated performance on the SKEMPI v2 dataset and compared against existing methods using rigorous cross-validation.

Main Results:

• DSSA-PPI demonstrated robust performance across diverse mutational contexts, outperforming existing methods on multiple benchmarks.
• The framework accurately identified mutations enhancing binding affinity in the SARS-CoV-2 receptor-binding motif (RBM)-ACE2 interaction.
• Guided optimization of a peptide inhibitor, achieving over 40-fold improvement in inhibitory activity against FXIa.

Conclusions:

• DSSA-PPI is a versatile and reliable tool for predicting mutation-induced perturbations in PPIs.
• The framework shows significant potential for advancing disease pathogenesis research and therapeutic development.
• Highlights the power of integrating structural and sequence data through advanced deep learning for biological predictions.