QSAR-Guided Discovery of Novel KRAS Inhibitors for Lung Cancer Therapy

Osasan Stephen Adebayo¹Oche Ambrose George^2,3*Daramola Olusola⁴Adefolalu Oluwafemi⁵Hind A. Alzahrani⁶Abdulkarim Hasan⁶

• ¹Prince Mishari Bin Saud Hospital, Baljurashi, Al-Baha, Al-Baha, Saudi Arabia
• ²Department of Plant Biology, Faculty of Life Sciences, University of Ilorin, Ilorin, Nigeria
• ³Centre for Malaria and Other Tropical Diseases Care University of Ilorin Teaching Hospital, Ilorin, Ilorin, Nigeria
• ⁴North Devon Council, Barnstaple, United Kingdom
• ⁵Obafemi Awolowo University Teaching Hospital Complex, Ife, Nigeria
• ⁶Albaha University, Al Aqiq, Saudi Arabia

KRAS mutations are a known driver of lung cancer, yet pharmacological targeting remains a challenge due to the protein's elusive binding sites. In this study, a quantitative structure–activity relationship (QSAR) modeling strategy was employed to predict the pIC₅₀ values of KRAS inhibitors and guide the discovery of novel lead compounds through de novo design. Chemopy was used to generate molecular descriptors of 62 KRAS inhibitors from the ChEMBL database (CHEMBL4354832). After applying dimensionality reduction and normalization, five machine learning techniques were used in developing the model: partial least squares (PLS), random forest (RF), stepwise multiple linear regression (MLR), GA-optimized multiple linear regression (GA-MLR), and XGBoost. R2, RMSE, and MAE were used in assessing performance. For feature interpretation, permutation-based and SHAP analyses were employed. PLS performed better than the other models (R2 = 0.851; RMSE = 0.292), and RF came in second (R2 = 0.796). Based on eight carefully chosen descriptors, the GA-MLR model demonstrated strong interpretability (R2 = 0.677). Virtual screening of 56 de novo–designed molecules using the GA-MLR model identified compound C9 (pIC₅₀ = 8.11) as a promising hit within the model's applicability domain. This integrative QSAR framework successfully identified novel KRAS inhibitor candidates with strong predicted potency.