Tailored molecular modelling and machine learning solutions for small-molecule drug discovery
Filella-Merce, I, Vilalta-Mor, J, Guallar, Víctor · UPCommons institutional repository (Universitat Politècnica de Catalunya) · 2023
Computational techniques can help speed up the drug discovery process while reducing its associated costs [1]. Thanks to the breakthrough of vast biological and chemical data, Machine Learning approaches started to rise. Particularly, generative modelling networks (GMN) have benefited from using large chemical datasets of molecules [2]. These datasets are required to successfully train GMNs, as they use them to learn how to generate chemically viable molecules. To further refine the model and obtain targetspecific molecules, GMNs can incorporate an additional specific set of molecules with demonstrated experimental activity with the target. Additionally, an active learning phase can be integrated into a GMN through an iterative process to improve the model's predictability. At this stage, molecular modelling (MM) methods can take part. As determined by MM descriptors, favorable molecules are included in the specific set for the next generation round, while unfavourable ones are discarded.