Structure-Based Design and Machine Learning-Driven Prioritization of EGFR Inhibitors

The Epidermal Growth Factor Receptor (EGFR) is a significant target in cancer therapy as it facilitates tumor proliferation and persistence. EGFR Tyrosine Kinase Inhibitors (TKIs) have been shown to be effective in clinical settings, but resistance mutations limit their long-term effectiveness. As a result, there remains a need for new and more effective EGFR inhibitors.

We synthesized a chemical library by adding 15 different R groups to a urea-aryl hydrazone core at the two para positions. This gave us 81 different structures. To determine pIC₅₀ values, molecular fingerprints were generated using PaDEL-Descriptor and subsequently used as input features for a pre-trained Random Forest regression model. Afterward, molecular docking and 100-nanosecond molecular dynamics simulations were used to assess the best candidates' ability to bind to the EGFR kinase structure and their stability.

The scaffold-based enumeration yielded 81 compounds with distinct structures. A machine learning approach with a random forest model identified potential candidates with pIC₅₀ values greater than 6, indicating active status. Docking studies have demonstrated the stability of compounds that form hydrogen-bond interactions and hydrophobic contacts, comparable to those of Erlotinib. Molecular dynamics simulations have been employed to validate that these candidate complexes maintain stable interactions during the simulation time.

The computational approach has identified potential lead molecules with stable binding poses and predicted their activity from inhibitory concentrations, yielding results comparable to those of approved drugs. The analysis based on selective halogenation substitution enhanced the stability of interactions, preserving the major hinge-region interaction as a hydrogen bond. The technique would offer a more comprehensive approach for prioritizing lead molecules and for refining new EGFR inhibitors.

This research demonstrates the integration of scaffold-based chemical design and machine learning. Atomistic simulations speed up the search for EGFR inhibitors. This process produces drug-like candidates that are expected to perform effectively and remain stable, facilitating additional testing and improving cancer research.