Cortex discovers new preclinical candidates against HIV

Our team trained our AI model on publicly available data reporting the ability of various chemical compounds to inhibit HIV-1's glycoprotein 41 (gp41), a part of the apparatus it uses to infect human cells.

We commissioned the NCMM (Center for Molecular Medicine Norway) and SINTEF, a large Norwegian research organization to proceed to an in vitro screen of gp41 inhibitors in a molecular assay. Among the small molecule libraries available for testing, we picked the one in which we predicted the most novel hits.

Prior to the experiment, we committed our ranked top 10% compounds for their predicted probability of inhibitory activity to SINTEF, which allowed them to independently validate the accuracy of our predictions.

The result: among the librarie's 1950 compounds, 22 hits were found in the library, matching our estimate of the total number of hits to be found.

Two of our top 3 picks were confirmed as inhibitors.
Our top 10% selection contained 21 of the 22 hits (the probability of this happening by chance is 1 in 10¹⁸)
Our overall prediction accuracy, as quantified by the ROC-AUC, was outstanding (96%)

Cortex is patenting 19 novel anti-HIV gp41 inhibitors.