Predicting Protein-RNA Binding Affinity Changes via Spatial Coupling-Aware State Space Modeling

The accurate prediction of how genetic mutations affect protein-RNA binding is vital for understanding disease mechanisms. Testing every possible mutation variant is too costly, prompting the development of computational methods capable of precisely estimating mutation-induced changes in binding affinity, known as Delta-DeltaG. Researchers have introduced a new deep learning technique called iSCALE, which offers an interpretable and adaptable approach to this challenge.

iSCALE integrates an implicit multiscale encoding strategy known as Spatial Coupling-Aware Ligand Encoding into a bidirectional state space modeling framework. This innovative combination enables iSCALE to uncover generalizable patterns in the multiscale coupling that significantly outperform existing methods in predicting not only protein-RNA binding Delta-DeltaG but also protein stability Delta-DeltaG and protein-protein binding Delta-DeltaG.

By examining the model's attention scores, researchers found that they correspond well with structural characteristics, providing further validation of the model's effectiveness. Moreover, iSCALE exhibits strong discriminative abilities when predicting closely related samples, such as complexes that contain the same mutation but differ in ligands or the same complex but differ in mutation sites.

In the end, iSCALE emerges as a valuable computational tool for large-scale prediction of protein-RNA binding Delta-DeltaG, pushing the boundaries of our understanding of mutation-induced pathological outcomes.