Dataset
6,114 unique short peptides combine 3,057 curated positives with 3,057 length-matched operational comparators.
Dataset and curationToxiCompass-Lite
Short-peptide toxicity screening
Submit one or many 4-50 aa peptide sequences. Stage 1 screens for toxin-like comparator support; passing rows continue to a ranked shortlist of 13 retained target/effect-family labels.
- Input
- FASTA, TXT, CSV
- Mode
- Sequence + ESM2
- Output
- Scores + top-3
Model ready
Version-locked public research service
- Web
- v0.1.0
- Inference bundle
- toxicompass-lite-v2-20260802
- Data snapshot
- 2026-05-22
Sequence input
Run a prediction job
Scientific basis
From curated evidence to a reviewable shortlist
Two-stage task
Binary screening stays separate from positive-only multi-label ranking across 13 retained labels.
Workflow detailsLite and Full
Lite supports rapid sequence-first use; Full adds multimodal input signals for diagnostic review.
Model architectureMolecular context
From venom source to folded toxin
Representative visual references situate short peptide toxins in their biological and structural context.
These reference images provide scientific context only. They are not model inputs, benchmark records, or predicted mechanisms.
Paper figure 1
Benchmark, data views, and model architecture
The R1 architecture figure connects the two-stage task and label hierarchy with the Lite sequence workflow and multimodal Full architecture.
Open full-resolution figure