Sudarko Sudarko 

Department of Chemistry, Faculty of Mathematics and Natural Sciences, University of Jember  

Rimba Candra Kristyono

Department of Chemistry, Faculty of Mathematics and Natural Sciences, University of Jember

Anak Agung Istri Ratnadewi

Department of Chemistry, Faculty of Mathematics and Natural Sciences, University of Jember

Wuryanti Handayani

Department of Chemistry, Faculty of Mathematics and Natural Sciences, University of Jember

DOI: https://doi.org/10.19184/ICL.v3i1.940

ABSTRACT 

QSAR modeling allows for predicting activity through quantitative relationships between molecular structure and activity. This research uses DEEPScreen, which is a development of QSAR for searching new drugs. This research leverages DEEPScreen-QSAR modeling to optimize the predictive power of machine learning algorithms on a dataset of 645 molecules from previous research. The optimized model achieves an accuracy of 0.7461 and precision of 0.8169, demonstrating its effectiveness in the virtual screening stage. The optimized DEEPscreen-QSAR model is used to screen approximately 1.9 million small molecules in the ChEMBL database, resulting in binary classification predictions of active (1) molecules as 781,213 and inactive (0) molecules as 1,133,325 (molecules with IC50 activity ≤10,000 nM are considered active). The active (1) molecules obtained are screened again to find molecules that can be absorbed by the body (orally) using Lipinski’s RO5 with 0 deviations, resulting in 557,428 active molecules that can be absorbed by the body. These screening results are validated using molecular docking methods by linking protein and ligand to determine Gibbs free energy (∆G) and interactions using PyRx, PyMOL, and Biovia Discovery Studio programs. Based on the results of this research, candidate CHEMBL3302655, DHFR inhibitors with codes CHEMBL1384989, CHEMBL1729486 are recommended.  

Keywords: DHFR, inhibitor, QSAR, virtual screening, docking molecular.