AstraPTM2: A Context-Aware PTM Prediction Model for Broad-Spectrum PTMs
Full-length, context-aware PTM prediction — calibrated for discovery. AstraPTM2 combines ESM-2 embeddings, AlphaFold2 structural features, and motif-aware transformers to deliver reliable calls on 39 PTM types, from phosphorylation to sumoylation — no sequence truncation required.
Why AstraPTM2?
AstraPTM2 runs a full-protein analysis in a single shot, scoring every residue for 39 possible modifications. Its calibrated design doesn’t just flag potential sites — it gives you confidence thresholds you can act on, revealing hidden PTM patterns and guiding downstream experiments.
Multi-Label Predictions
AstraPTM2 detects whether a residue is modified and classifies its type from 39 PTM labels — in a single, unified framework, in a single pass.
Context-Aware
Proteins are complex, and context is crucial. AstraPTM2 handles proteins of virtually any length (no windowing or truncation), preserving long-range dependencies.
Discovery-Oriented
By leveraging robust context-awareness, a large training dataset, and dual modes, AstraPTM2 allows users to separate high-confidence actionable calls from hypothesis-generating lower-confidence sites.
Proven Performance Across 39 PTMs
AstraPTM2 achieves macro-F1=59% and AUROC=99% across 39 PTM types — from phosphorylation and ubiquitination to rare labels like lactylation. Calibrated outputs deliver actionable, residue-level predictions ready for experimental use.
Next-Generation Architecture
Built on a transformer backbone with multi-scale convolutions and global gating, AstraPTM2 captures both local motifs and global context. Trained on 100k+ proteins with ESM2 embeddings and AlphaFold2 features, it generalizes across common and rare PTMs.
See AstraPTM2 in action.
Curious how AstraPTM2 streamlines PTM identification from a simple sequence input? Book a demo, and let us showcase you how to harness AstraPTM2's capabilities for your research.
