SAPP: Structure Aware PTM Prediction
Year: 2025
Type: article
Abstract: Post-translational modifications (PTMs) play critical roles in regulating cellular processes such as signal transduction, cell growth, and differentiation. Accurate identification of PTM sites is fundamental to understanding cellular mechanisms and developing therapeutic interventions. However, traditional computational models have predominantly relied on sequence data alone, neglecting important structural contexts such as intrinsically disordered regions and solvent accessibility. To address this gap, we introduce SAPP (Structure-Aware PTM Prediction), a pioneering model that integrates structural features derived from AlphaFold2 predictions with sequence information using a unified Transformer-based framework. Utilizing self-attention and cross-attention mechanisms, SAPP effectively captures complex interactions between sequences and their structural states, improving prediction accuracy and biological relevance over sequence-based models. Notably, SAPP is among the first structure-based PTM prediction frameworks, which allows for fine-tuning from a phosphorylation-pretrained model to other PTM types, achieving generalization performance in PTM types with limited training data. This supports the critical role of structural information in PTM prediction, deepening our understanding of their biological significance. • SAPP (Structure-Aware PTM Prediction) predicts PTM sites using 3D protein structures. • Structural features improve prediction accuracy beyond sequence motif patterns. • SAPP performance varies with structure quality, suggesting future model improvements. • Transfer learning from a phosphorylation model outperforms training from scratch.
Authors: Yujin Choo, Seungjin Na, Eunok Paek
Institutions: Hanyang University, Korea Basic Science Institute
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: hybrid
Estimated APC paid: $3,080
OpenAlex ID: W4414933769