Download
Everything behind this site is public. The code is on GitHub, the reference data are fetched by one command, and the index and the checkpoint are published by the sites that maintain them.
The code
github.com/smuskal/toxpred. The package this site imports. The consortium vote, the reach calculation and the contribution formats are all in it, and every number in the paper was produced by the code it wraps. The site does not reimplement any of it, so the site and the command line answer the same question the same way.
git clone https://github.com/smuskal/toxpred.git
cd toxpred
conda env create -f environment.yml
conda activate toxpred
toxpred fetch --with-reach
toxpred score --smiles 'CC(=O)Oc1ccccc1C(=O)O' --reachtoxpred fetch downloads the first three rows below into ~/.toxpred, and --reach adds the checkpoint. Each one records the URL, the size and the sha256 of the file it took, which toxpred provenance prints back. The family assignments in the last row are passed on the command line, which the row says how to do.
The reference data
| Data | Where it comes from | What it is |
|---|---|---|
| CATMoS, rat oral LD50 | NIEHS OPERA | Experimental values only. 8,599 compounds in the sampled reference set, 5,953 of them carrying an exact value and able to vote. |
| TOXRIC, the 28 further endpoints | TOXRIC | The endpoint measurements behind every row of the endpoints table. |
| The Reverse Screen index, release 2026-10-29 | reversescreen.ai | 33,642 indexed ligands and the protein sites they were solved against. The second signal is computed against this. |
| The PharmCast checkpoint | pharmcast.ai | pharmcast_scp_v10.pt, the published checkpoint, which is the one the index was built with and the only one this site uses. |
| Protein family assignments | familyfoundationmodel.com | The family each target belongs to. Hand it to toxpred score as --family-map and reach is counted in families; on its own the command counts the proteins reached, which is the other column of the same screen. |
Checksums
The exact files this site is serving against. Verify a download against the digest beside it.
| File | Bytes | SHA256 |
|---|---|---|
| catmos_experimental.csv | 1,795,064 | 021f8c1f9eab38d1df87caaaf4537c5b144d99682b6c763058fd2ac489b1e38d |
| reverse_screen_index_2026-10-29.pfp.gz | 13,293,185 | a701ade1fbc2e19ac17e5fd0291a41ebeed012f04868f88aa8126531c84c3784 |
| reverse_screen_ligands_2026-10-29.tsv | 2,278,616 | 6cfacd6b78c82f4afac805ba755ec92d3902d522d90c082fc878afbdcfe78750 |
| reverse_screen_sites_2026-10-29.tsv | 3,657,646 | a7fddb8007d0937995cad7c8e72d1d9ea561f1d1bad95e88809c314f4ff9e5a7 |
| reverse_screen_targets_2026-10-29.tsv | 5,147,292 | fc445103c555460e3d8d4fb4b23ffdc9ab7b8ee73f649ecf2bcea98129ce31a6 |
The checkpoint is published with its own SHA256SUMS at pharmcast.ai, and toxpred verifies it on download rather than trusting the transfer.
The paper
Muskal SM, Jha SK, Kishore P, Tyagi P. Toxicity by Consortium Revisited: Structural Neighbors and Cross-Family Reach. bioRxiv 2026. doi:10.64898/2026.10.04.756566