E-pRSA

E-pRSA predicts Relative Solvent Accessibility (RSA) of residues directly from protein sequences using protein language model embeddings to enable structure-independent analysis of protein surface exposure and functional sites.

Key Features:

• Protein Sequence-Based Predictions: E-pRSA estimates RSA values directly from protein sequences without requiring three-dimensional structures or multiple sequence alignments.
• Utilization of Protein Language Models: It employs two complementary Protein Language Models to generate embeddings that facilitate RSA prediction, replacing sequence-profile-based approaches such as DeepREx that relied on multiple sequence alignment-derived profiles.
• Benchmarking Performance: E-pRSA demonstrated state-of-the-art performance on blind test sets, outperforming DeepREx in RSA prediction benchmarks.

Scientific Applications:

• Protein-Protein Interactions: Predicted RSA values help identify potential interacting surfaces and interfaces on proteins.
• Characterization of Variations: RSA predictions support analysis of sequence variations and mutations that may affect protein function or stability.

Methodology:

E-pRSA generates embeddings from two complementary Protein Language Models and predicts residue-level RSA directly from those embeddings without using experimental 3D structures or multiple sequence alignments.