Astra Suite

AstraPTM2

Post-Translational Modifications, Residue-by-Residue

Proteins are rarely just their sequence — they're phosphorylated, glycosylated, ubiquitinated and more. AstraPTM2 reads a sequence together with its predicted structure and flags where 39 different modification types are likely, one residue at a time.

Per-Residue39 PTM TypesSequence + Structure
Read the Preprint
M
N
G
G
P
T
E
G
P
N
F
Y
S
V
P
F
P
S
N
K
P PhosphorylationS DisulfideG N-Glycosylation
Per-Residue PTM Calls · N-Terminus Of Rhodopsin (P08100)
Signature Strength

The Broadest PTM Coverage in One Model

39 modification types — from the common regulatory marks to rare lipidations and processing events — scored at single-residue resolution across a full-length protein of any length, with no windowing or truncation.

0.91
F1 on the Best Resolved Marks

Real Structure · Ubiquitin (PDB 1UBQ)

Modifications, mapped onto the fold

What It Predicts

39 Modification Types, One Model

From the common regulatory marks to rare lipidations and processing events — each scored per residue, so you know not just whether a protein is modified, but where.

PhosphorylationAcetylationUbiquitinationDisulfide BondN-Linked GlycosylationO-Linked GlycosylationC-Linked GlycosylationSUMOylationMethylationS-PalmitoylationN-PalmitoylationMyristoylationFarnesylationGeranylgeranylationGPI-AnchorHydroxylationSulfationNitrationS-NitrosylationCitrullinationSuccinylationMalonylationGlutarylationCrotonylationButyrylationLactylationADP-RibosylationNeddylationAmidationDeamidationOxidationSulfoxidationFormylationγ-CarboxyglutamatePyroglutamateGlutathionylationDephosphorylationLactoylationS-Diacylglycerol

Performance

Common Marks, Called Reliably

Coverage is broad — 39 types in a single pass — and the common regulatory marks, where public data is richest, are called most reliably.

0.59
Macro-F1 Across All 39 Types
Held-Out Test
0.91
F1 on the Best Resolved Marks
Common PTMs
39
Modification Types, One Pass
Single-Residue
Per-type recognition
Phosphorylation0.86
Disulfide Bond0.82
N-Linked Glycosylation0.75
Acetylation0.71
Ubiquitination0.63
SUMOylation0.41

Per-type F1, representative. Macro-F1 across all 39 types is 0.59 on held-out test — the common regulatory marks score above 0.7; rare modifications pull the average down. Full benchmarks are in the preprint.

In Context

Mapped Onto the Fold

Sites don't live in a vacuum. AstraPTM2's calls are shown against the structure and membrane topology, so a predicted glycosylation on an extracellular loop or a disulfide across a pocket reads at a glance.

EXTRACELLULARCYTOPLASMTM1TM2TM3TM4TM5TM6TM7ICL1ECL1ICL2ECL2ICL3ECL3GPPNCPTMBindingDisulfideTM helixpLDDT
Predicted PTM Sites & Topology · Bovine Rhodopsin (UniProt P08100)

How It Reads

Sequence and Structure, Together

Sequence Context

A protein language model captures the local and long-range context around every residue.

Structural Context

Predicted structure features — confidence, exposure, secondary structure — tell modification-likely surfaces apart.

Per-Residue Calls

Each residue gets a calibrated probability for every modification type, ready to overlay on the sequence.

AstraPTM2: A Context-Aware Transformer for Broad-Spectrum PTM Prediction
Methods, coverage, and evaluation against public baselines · bioRxiv, 2025
Read →

Put Astra on Your Targets

Every Astra model runs inside the Orbion platform. Bring a sequence and get the full read-out — structure, function, modifications, binding, and stability.