Learning the PTM code through a coarse-to-fine mechanism-aware framework

2026-05-15 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning · Depth: Expert, short

Summary

COMPASS-PTM is a novel, mechanism-aware, coarse-to-fine learning framework designed to decipher the complex combinatorial "code" of post-translational modifications (PTMs). This framework unifies residue-level multi-label PTM prediction with enzyme-substrate assignment by jointly modeling PTM patterns and their catalytic regulators. Built upon protein language models, COMPASS-PTM integrates physicochemical descriptors and a crosstalk-aware prompting mechanism to learn biologically coherent patterns of cooperative and antagonistic modifications, while also addressing the dual long-tail distribution inherent in PTM data. The model significantly outperforms existing baselines across multiple proteome-scale benchmarks, achieving a 122% relative improvement in F1-score for multi-label site prediction and a 54% gain in zero-shot enzyme assignment. Furthermore, COMPASS-PTM demonstrates interpretable generalization, recovering canonical kinase motifs and linking missense variants to PTM disruptions and enzyme-substrate network rewiring.

Key takeaway

For AI Scientists and Research Scientists working on protein function and cellular signaling, COMPASS-PTM offers a robust framework to simultaneously predict PTM sites and their regulatory enzymes. You should consider integrating this mechanism-aware, coarse-to-fine learning approach to improve the accuracy and interpretability of your PTM analyses, especially when dealing with complex combinatorial codes and long-tail data distributions. This could lead to more precise understanding of protein regulation and disease mechanisms.

Key insights

COMPASS-PTM unifies PTM site prediction and enzyme assignment by modeling PTM patterns and catalytic regulators.

Principles

• Integrate physicochemical descriptors for biological coherence.
• Address dual long-tail distributions in PTM data.
• Couple statistical learning with explicit biochemical knowledge.

Method

COMPASS-PTM uses a coarse-to-fine learning framework, building on protein language models, integrating physicochemical descriptors and a crosstalk-aware prompting mechanism to jointly model PTM patterns and enzyme-substrate assignments.

In practice

• Predict multi-label PTM sites with high accuracy.
• Perform zero-shot enzyme-substrate assignment.
• Interpret mechanistic links of missense variants.

Topics

• Post-translational Modifications
• COMPASS-PTM
• Enzyme-Substrate Assignment
• Protein Language Models
• Multi-label PTM Prediction

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Reactive Flux Matching: Mechanism Discovery and Adaptive Sampling of Rare Events — Flux Matching learns reaction pathways and coordinates directly from reactive trajectory data without prior dynamic knowledge.
• Biological Reasoning-Informed Regression for Interpretable Regulatory DNA Activity Prediction — Integrating structured biological knowledge with LLMs via two-stage training improves interpretable DNA regulatory activity prediction.
• A large-scale unified deep learning model for peptide mass spectrum interpretation trained on multimodal data — pUniFind unifies peptide-spectrum scoring and de novo sequencing, significantly improving proteomic analysis sensitivity and modification coverage.
• Move over, AlphaFold: open-source model predicts shape of 1 billion proteins — ESMFold2, an open-source AI, predicts over a billion protein structures, outperforming AlphaFold3 and enabling novel protein design.
• ChemAmp: Amplified Chemistry Tools via Composable Agents — ChemHAS uses hierarchical agent stacking to reduce prediction errors in chemistry tools, outperforming existing LLM and specialist models.
• EnzymeTuning improves enzyme-constrained metabolic modeling and proteome abundance prediction through deep learning — EnzymeTuning, a GAN-based framework, optimizes *k*cat values to enhance metabolic modeling and proteome prediction accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.