Graphical representation of biomolecules

Abstract

In an effort to enhance the process of drug discovery and development, this work explores the possibility of creating an extensive and useful substructural-based chemical fingerprint for virtual drug screening by using differently sized subgraphs generated from biomolecule-converted graphs as representative features. The experiment was done using data from ChEMBL, an open-source chemical library, in the context of the BRAF protein and ligand, where the goal was to classify whether a biomolecule is a BRAF ligand, with features being substructures within the biomolecule. The effectiveness of the representation is evaluated through a classification task, where six models were constructed, trained, and tested using the newly created representation. It was ultimately concluded that using size 4 subgraphs as features produced the best results, and that the new representation does have the potential to be used for virtual drug screening.