SA-MTP framework accurately annotates multifunctional therapeutic peptides by integrating structural and sequence AI.

Therapeutic peptides are promising drug candidates, offering high target specificity and favorable safety profiles, bridging the gap between small molecules and biologics. However, their accurate functional annotation remains a significant challenge due to their short sequence length, strong structural flexibility, and the common presence of multiple biological functions within a single peptide. Existing computational methods often struggle to capture these complex characteristics, hindering efficient drug discovery and development. A robust framework is needed to overcome these limitations and accelerate the identification of novel peptide therapeutics.

Researchers developed SA-MTP, a structure-aware framework for multifunctional therapeutic peptide annotation. This framework integrates pretrained protein language models to capture deep sequence semantics with a graph attention network designed to process probabilistic structural features. To account for the inherent conformational variation in short peptides, SA-MTP constructs input-dependent structure-aware graphs. The framework's performance was rigorously benchmarked across 15 therapeutic function categories using comprehensive datasets, comparing its predictive capabilities against existing annotation methods.

The framework's ability to integrate sequence semantics from pretrained protein language models with probabilistic structural features via a graph attention network proved critical. By constructing input-dependent structure-aware graphs, SA-MTP effectively modeled the conformational variations characteristic of short peptides, which is a key factor in their diverse biological activities. This comprehensive approach allowed for more precise and reliable predictions across the 15 distinct therapeutic function categories evaluated. The consistent improvement across various metrics highlights the framework's robustness and its potential to overcome current annotation challenges.

SA-MTP achieved superior performance compared to existing methods across multiple evaluation metrics, including accuracy, F1-score, and Matthews correlation coefficient, demonstrating its enhanced capability in multifunctional therapeutic peptide annotation.