Machine Learning-Based Virtual Screening and Identification of the Fourth-Generation EGFR Inhibitors

Epidermal growth
factor receptor (EGFR) plays a pivotal regulatory
role in treating patients with advanced nonsmall cell lung cancer
(NSCLC). Following the emergence of the EGFR tertiary CIS C797S mutation,
all types of inhibitors lose their inhibitory activity, necessitating
the urgent development of new inhibitors. Computer systems employ
machine learning methods to process substantial volumes of data and
construct models that enable more accurate predictions of the outcomes
of new inputs. The purpose of this article is to uncover innovative
fourth-generation epidermal growth factor receptor tyrosine kinase
inhibitors (EGFR-TKIs) with the aid of machine learning techniques.
The paper’s data set was high-dimensional and sparse, encompassing
both structured and unstructured descriptors. To address this considerable
challenge, we introduced a fusion framework to select critical molecule
descriptors by integrating the full quadratic effect model and the
Lasso model. Based on structural descriptors obtained from the full
quadratic effect model, we conceived and synthesized a variety of
small-molecule inhibitors. These inhibitors demonstrated potent inhibitory
effects on the two mutated kinases L858R/T790M/C797S and Del19/T790M/C797S.
Moreover, we applied our model to virtual screening, successfully
identifying four hit compounds. We have evaluated these hit ADME characteristics
and look forward to conducting activity evaluations on them in the
future to discover a new generation of EGFR-TKI.