# PharmCast

> PharmCast predicts a complete three-dimensional pharmacophore fingerprint,
> 10,549 bits wide, directly from a SMILES string, so a pharmacophoric
> similarity comparison runs without generating a conformer ensemble. Built by
> Eidogen-Sertanty, Inc. Principal author Steven M. Muskal, ORCID
> 0000-0002-3487-270X (https://orcid.org/0000-0002-3487-270X).

The fingerprint is the three-point PharmPrint encoding: every bit is one
triangle of three pharmacophore features at three binned distances. Comparing
two of them is PharmSim. The conventional route to that fingerprint spends
almost all of its time generating conformers; PharmCast removes that stage
rather than accelerating it.

Version 10 is the current release, trained on 5,887,229 molecules drawn from a
screening collection, activity-backed ChEMBL compounds and peptide loops taken
from protein structures. Agreement with the reference calculation is Pearson
0.980 on screening collection chemistry.

## Pages

- [PharmCast](https://pharmcast.ai/): what the model is, how it is trained and tested, and where it is weakest
- [Compare two molecules](https://pharmcast.ai/compare.html): draw or paste two structures and get a pharmacophoric similarity
- [Fingerprint a molecule](https://pharmcast.ai/fingerprint.html): one structure to its predicted fingerprint
- [Model weights](https://pharmcast.ai/models): released checkpoints with SHA-256 checksums and the training snapshot each was built from

## Citation

Muskal, S. M. and McGregor, M. J. PharmCast: rapid generation of
three-dimensional pharmacophore fingerprints from two-dimensional structure
without conformer generation. bioRxiv, 2026.
doi:10.64898/2026.09.02.748999 (https://www.biorxiv.org/content/10.64898/2026.09.02.748999v2)

Authors: Steven M. Muskal (https://www.stevenmuskal.com/) and Malcolm J. McGregor.

## Related

- [Eidogen-Sertanty](https://eidogen-sertanty.com/): the company
- [How pharmacophore fingerprints work](https://eidogen-sertanty.com/pfp.php): the method page
- [Reverse peptide mimetics](https://eidogen-sertanty.com/reversePeptideMimetic.php): PharmCast fingerprints used to find peptides that present the same pharmacophore as co-crystal ligands
- [Source code](https://github.com/smuskal/pharmcast): Apache-2.0, code and model utilities
- [Reverse Screen](https://reversescreen.ai/): built on PharmCast. Every co-crystal ligand in the Protein Data Bank is indexed by the fingerprint PharmCast predicts, so a query molecule retrieves candidate targets in milliseconds and docking runs only into what retrieval returns. Paper: Muskal, S. M.; Nicola, G. bioRxiv 2026, doi:10.64898/2026.09.09.750461 (https://www.biorxiv.org/content/10.64898/2026.09.09.750461v1). Video: https://www.youtube.com/watch?v=mFRBfK6iv_w
- [Family Foundation Model](https://familyfoundationmodel.com/), [GPCR Foundation Model](https://gpcrfoundationmodel.com/) and [Kinase Foundation Model](https://kinasefoundationmodel.com/): comparator models from the same group that rank which compound a target prefers and which target a compound prefers. All three on one page: https://eidogen-sertanty.com/foundationmodels.php. Video overview: https://www.youtube.com/watch?v=VWMWP5hi3uA
- [Steven M. Muskal](https://www.stevenmuskal.com/): principal author. PharmCast for a general reader, with the paper: https://www.stevenmuskal.com/projects.html#pharmcast. Publications: https://www.stevenmuskal.com/publications.html

## What this is not

PharmCast predicts what a molecule can present in three dimensions. A high
pharmacophoric similarity is not a measurement of biological activity, and the
model returns an ensemble fingerprint rather than a specific conformer.