Creating an efficient screening model for TRPV1 agonists using conformal prediction

Members of the transient receptor potential (TRP) family, such as the vanilloid receptor type 1 (TRPV1), have been shown to play a significant role in nociception. Activation of TRPV1 not only causes pain, but can also lead to inflammation in the surrounding tissues through the induced influx of primarily Ca2+ ions into the cytosol. Therefore agonistic properties of a compound towards TRPV1 might indicate harmful environmental stimuli, which can be screened for using various low and high throughput in vitro methods. Prediction of such properties for novel entities using existing data is of importance, however the generation of in silico models are usually difficult due to the high class imbalance ratio present within these datasets. This imbalance can be quite significant, such as in the case of PubChem BioAssay AID 540275, with the inactive:active ratio of 484:1. In this study, we propose a first-tier high throughput in silico screening approach using the Mondrian Conformal Prediction framework, to identify potential agonists of TRPV1 from 2D molecular structures using RDKit descriptors in the PubChem BioAssay AID 540275 dataset.