PharmCast: rapid generation of three-dimensional pharmacophore fingerprints from two-dimensional structure without conformer generation
A three-dimensional pharmacophore fingerprint records the binding features a molecule can present. It is a description of a hand in search of a glove. Because it is defined by presented features instead of two-dimensional structure, it can identify pharmacophoric similarity between structurally distinct compounds and support scaffold hopping and the identification of structurally distinct…
PharmCast is a groundbreaking tool that can generate three-dimensional pharmacophore fingerprints from a molecule's two-dimensional structure, all without the need for conformer generation. This innovative approach allows for the identification of pharmacophoric similarity between structurally distinct compounds, making it easier to support scaffold hopping and find structurally distinct compounds with comparable binding features.
Traditionally, generating a pharmacophore fingerprint required predicting the binding features of a molecule through conformer generation, which was a time-consuming and costly process. However, with PharmCast, the conformational stage has been removed, and the feedforward neural network directly predicts all 10,549 bits of a PharmPrint ensemble fingerprint from a SMILES string.
This results in a significant speed-up in the fingerprint generation process, with PharmCast generating fingerprints for two molecules and comparing them in just 0.584 ms, while the conventional conformer-based pipeline took 5.71 s.
PharmCast was trained on a vast dataset of 5,887,229 molecules from a screening collection, activity-backed ChEMBL compounds ranging from 142 to 1000 Da, and peptide loops excised from crystal structures. When evaluating PharmCast on 155,648 purchasable catalog compounds, 139,700 activity-backed ChEMBL compounds, and 13,500 peptide loops reserved for testing, the median fingerprint error was 0.008 for screening collection chemistry, 0.016 for loop peptides, and 0.027 for activity-backed ChEMBL compounds.
The corresponding Pearson r values were 0.980, 0.984, and 0.936, and the pairwise ranking accuracy was 0.936 for screening collection chemistry, 0.952 for loop peptides, and 0.889 for activity-backed ChEMBL compounds. The reference calculation reproduced itself with an error of 0.006 and r of 0.995.
By predicting ensemble pharmacophore fingerprints from constitution alone, PharmCast offers a cost-effective solution for screening collections and optimizing molecules, making it a valuable tool in the realm of virtual screening and scaffold hopping.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
