Abstract
Effective inhibitors for the epidermal growth factor receptor (EGFR) still present a major obstacle in the development of cancer drugs. In this work, we create new EGFR inhibitors using generative artificial intelligence (AI) by fine-tuning a GPT-2 model on a dataset comprising around 500,000 molecules from the ChEMBL database. We evaluate our method against a Long Short-Term Memory (LSTM) network trained on the same dataset to create benchmarks. Before being filtered depending on Lipinski’s rule of five, synthetic accessibility, and drug-likeness scores, the produced compounds were evaluated for validity, distinctiveness, and novelty. To assess their binding affinities, chosen candidates underwent molecular docking experiments against EGFR (PDB ID: 1M17). Our results show that although GPT-2 excels in producing structurally varied molecules, the LSTM model generates a larger fraction of chemically valid compounds. Many produced candidates showed good binding interactions with EGFR, therefore highlighting the possibilities of deep learning-based generative models in hastening the early phases of therapeutic development. This work presents a scalable method for creating tailored cancer treatments by showing the synergy between AI and in silico drug design.
