HieraPTM: From PTM Site Recognition to Functional Phosphosite Prediction through Hierarchical Context and Protein Mechanics

Protein post-translational modifications (PTMs) regulate cellular signaling and protein function, making accurate site identification essential for understanding protein regulation. However, PTM site recognition and functional phosphosite prediction are largely modeled independently, limiting the transfer of large-scale recognition knowledge to functional prioritization. Here, we introduce HieraPTM-HieraFuncPhos, a cross-task hierarchical framework that separately models and conditionally integrates transferable PTM recognition representations and function-specific mechanical evidence. HieraPTM integrates local sequence patterns, protein language model representations, and three-dimensional structural information to recognize multiple PTM types. HieraFuncPhos transfers these representations through parameter-efficient adaptation and adaptively retrieves local and distal mechanical evidence conditioned on functional representations. At HSP90AA1 T36 and T115, the model reveals distinct mechanical response patterns consistent with published nuclear magnetic resonance (NMR) observations, suggesting that the framework not only prioritizes functional phosphosites but also provides biophysically grounded, testable hypotheses regarding site-specific regulation. HieraPTM achieved macro-averaged AUROC and AUPRC values of 0.9237 and 0.8425 across eight PTM tasks, while maintaining competitive performance on severely imbalanced external benchmarks. HieraFuncPhos achieved an AUROC of 0.8274 and an AUPRC of 0.8188, outperforming existing comparison methods. Together, this framework connects PTM site recognition, functional prioritization, and structure-informed hypotheses regarding potential regulatory processes.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-13
DOI
https://doi.org/10.64898/2026.11.12.766127
Primary Topic
Machine Learning in Bioinformatics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

HieraPTM: From PTM Site Recognition to Functional Phosphosite Prediction through Hierarchical Context and Protein Mechanics

林婵微, Tianchi Lu, Yanze Niu, Hao Li et al.
2 links
bioRxiv (Cold Spring Harbor Laboratory)
Machine Learning in Bioinformatics
preprint

HieraPTM: From PTM Site Recognition to Functional Phosphosite Prediction through Hierarchical Context and Protein Mechanics

林婵微, Tianchi Lu, Yanze Niu, Hao Li, Tianhang Chen, Di Wang, Changwei Lin, Pengyu Chen, Yuzhi Yuan, Hengjia Zhang, Ziwei Su, Xicheng He, Zhiyuan Yang
preprint en

Abstract

Protein post-translational modifications (PTMs) regulate cellular signaling and protein function, making accurate site identification essential for understanding protein regulation. However, PTM site recognition and functional phosphosite prediction are largely modeled independently, limiting the transfer of large-scale recognition knowledge to functional prioritization. Here, we introduce HieraPTM-HieraFuncPhos, a cross-task hierarchical framework that separately models and conditionally integrates transferable PTM recognition representations and function-specific mechanical evidence. HieraPTM integrates local sequence patterns, protein language model representations, and three-dimensional structural information to recognize multiple PTM types. HieraFuncPhos transfers these representations through parameter-efficient adaptation and adaptively retrieves local and distal mechanical evidence conditioned on functional representations. At HSP90AA1 T36 and T115, the model reveals distinct mechanical response patterns consistent with published nuclear magnetic resonance (NMR) observations, suggesting that the framework not only prioritizes functional phosphosites but also provides biophysically grounded, testable hypotheses regarding site-specific regulation. HieraPTM achieved macro-averaged AUROC and AUPRC values of 0.9237 and 0.8425 across eight PTM tasks, while maintaining competitive performance on severely imbalanced external benchmarks. HieraFuncPhos achieved an AUROC of 0.8274 and an AUPRC of 0.8188, outperforming existing comparison methods. Together, this framework connects PTM site recognition, functional prioritization, and structure-informed hypotheses regarding potential regulatory processes.

bioRxiv (Cold Spring Harbor Laboratory)
Hong Kong Polytechnic University (HK), City University of Hong Kong (HK), Lanzhou University (CN)
Machine Learning in Bioinformatics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon