We have currently released a version of our notebook platform to demonstrate how PSICHIC_XL (still known as Pre-trained PSICHIC in our preprint and in the 'trained_weights' folder of this repo) can be applied to the crucial task of selectivity profiling. This process is vital post-hit identification to ensure the efficacy of small-molecule ligands and to reduce off-target effects, a key consideration in drug discovery due to the polypharmacology inherent in many compounds. Our case study on AR subtypes illustrates PSICHIC_XL's capability to predict ligand promiscuity or selectivity across the highly conserved orthosteric binding sites of the four adenosine receptor (AR) subtypes: A₁R, A_2AR, A_2BR, and A₃R. It further demonstrates how interaction fingerprints can interpret the mechanism of interactions, identifying selectivity determinants directly from sequence data—a particularly challenging task for A₁R due to the similarity across AR subtypes.

We seek to further develop a webserver version for selectivity profiling in the future to make it widely accessible. In the meantime, we invite you to explore the current version, adaptable for any protein dataset: Colab Online Platform

Stay tuned for the full release of our platform, set to provide insights into achieving subtype selectivity, a crucial step towards the development of safer and more effective therapeutics.