A generative model for the design of novel inhibitors targeting the PD-1/PD-L1 pathway.

2576 Background: The programmed cell death-1/programmed cell death ligand-1 (PD-1/PD-L1) signaling pathway plays a pivotal role in tumor immunosuppression. However, the design of de novo molecules with precise pharmacological and molecular properties remains a resource-intensive and financially demanding endeavor. It is hypothesized that generative models trained on molecular graph encodings can design novel inhibitors targeting the PD-1/PD-L1 pathway. Objective: This study aims to develop a generative model capable of designing novel, orally bioavailable inhibitors of the PD-1/PD-L1 pathway. Methods: A large language model was pre-trained on 1 million chemical structures derived from the ChEMBL database. Each structure was represented using the Simplified Molecular Input Line Entry System (SMILES) strings, which were further tokenized into discrete atomic and functional group-level tokens. The model employs an Average-Stochastic Gradient Descent Weight-Dropped Long Short-Term Memory (AWD-LSTM) architecture. Transfer learning was applied to fine-tune the pre-trained language model on the target chemical structures, enabling domain-specific adaptation for the desired application space. Results: The model demonstrated robust performance in generating chemically valid, unique, and novel inhibitors targeting the PD-1/PD-L1 pathway. It achieved a validity rate of 97%, a uniqueness rate of 96%, a novelty rate of 95%, and a diversity score of 76.04%. Additionally, the generated molecules exhibited favorable physicochemical properties, including a logarithm of the partition coefficient (LogP) of 4.52, atopological polar surface area (TPSA) of 113.06 Angstrom squared, an average of 9.62 rotatable bonds, 2.77 hydrogen bond donors, and 6.79 hydrogen bond acceptors . Conclusions: A generative model was developed to design novel, orally bioavailable inhibitors of the PD-1/PD-L1 pathway. This approach provides an efficient and automated tool for designing de novo molecules with precise molecular and pharmacological properties, potentially accelerating drug discovery in immuno-oncology.