Resources
The public project repository provides versioned code, benchmark records, checksummed release assets, and reproduction guidance. A permanent DOI has not been assigned.
Downloads
Public GitHub repository cited in the R1 manuscript.
Code, model/data archives, checksums, examples, and reproducibility records under tag benchmark-v1.0.0.
Revision-period Stage 1 and Stage 2 counts, metrics, and interpretation limits.
Frozen aggregate values and source-artifact hashes; no row-level source sequences.
Checksums for public files plus the source citation and redistribution boundary.
R1 manuscript Figure 1, PNG.
R1 manuscript Figure 2, PNG.
R1 manuscript Figure 3, PNG.
R1 manuscript Figure 5, PNG.
Paper figure 3
Component-matched ablations, first-reference-label rank composition, and score distributions help show where model components change ranking behavior. These analyses do not turn scores into biological certainty.
Open full-resolution figure
Release status
The frozen benchmark release is published with immutable provenance. The DOI remains unassigned, so cite the release tag and commit shown below.
Service continuity
This website is a version-locked research service, not the sole record of the study. It exposes the sequence-only Lite workflow; database MSA, Protenix, GVP, and Full-mode analyses remain offline research workflows.
Public HTTPS service: https://lmmd.ecust.edu.cn/toxicompass/
87805bef0f652f89cae9dc2739f64c29d49f4f3fAvailability and security fixes may update the web layer without changing scores. New data, splits, thresholds, or model weights receive a new release identity and changelog; prior versions remain archived and result exports retain their original provenance.
LMMD commits to maintain the service for at least 3 years after article publication. Professor Yun Tang (ytang234@ecust.edu.cn) is the maintenance contact. Compatibility, dependency, and release status are reviewed at least annually; uninterrupted uptime is not guaranteed.
Planned retirement will be announced on the website and repository at least 12 months in advance, with the final supported release identified. Source, benchmark records, checksums, and prior tagged versions remain available; an urgent security or infrastructure suspension may occur sooner and will be documented promptly.
fc712da6cb1f11c6f0ea253eb40dff20a485666fd148dc40fb6c5e5d59ae0ce3443e804551a2cafcff1a6e3162a724fbcb42eef400b20793f50d101b16bad6691094d7bdabd38e60619f45ed70055e1d3c279d43df07820a4cac2e08500fc0cdd4a7526927fa15031026841a1feabab89bb3db6329732153c1b4d36df807ac48Compute envelope
Limits keep anonymous GPU-backed screening responsive and make the distinction from the Full research workflow explicit.
Benchmark source directory
The frozen benchmark records retain their original source provenance. Use the source services for database browsing and citation; use the tagged ToxiCompass release for the exact normalized benchmark snapshot and split records.
Core references
BioTDWang et al., Journal of Chemical Information and Modeling, 2026.Database website
UniProtThe UniProt Consortium, Nucleic Acids Research, 2025.Knowledgebase website
MMseqs2Steinegger & Söding, Nature Biotechnology, 2017.
ESM2Lin et al., Science, 2023.
GVPJing et al., ICLR, 2021.
MMoEMa et al., KDD, 2018.
Visual sources
Conotoxin GS, PDB 1AG7 Solution NMR ensemble. Public-domain image by Jawahar Swaminathan and MSD staff at the European Bioinformatics Institute. Image record
Beta-2-bungarotoxin, PDB 1BUN Crystal structure. Public-domain image by Jawahar Swaminathan and MSD staff at the European Bioinformatics Institute. Image record
Conus textile Photograph by Richard Ling, licensed under CC BY-SA 3.0. Image and license
Post-hoc robustness sources
ConoServer Conopeptide candidates. Cite ConoServer; its database and tools are limited to academic research unless other permission is obtained. Source website
ArachnoServer Spider-toxin challenge candidates. Row-level records are not mirrored because no blanket redistribution grant was identified. Source website
Hemolytik 2.0 Assay-scoped hemolytic and non-hemolytic evidence. Cite Singh et al. (2026), DOI 10.1021/acs.chemrestox.5c00322. Source website
IUPHAR/BPS Guide to PHARMACOLOGY Therapeutic-peptide and disulfide-rich comparator evidence. Database ODbL; contents CC BY-SA 4.0. Source website
