Advancement in In-Silico Drug Discovery from Virtual Screening Molecular Dockings to De-Novo Drug Design Transformer-based Generative AI and Reinforcement Learning

Author

Dony Ang, Chapman UniversityFollow

Date of Award

8-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational and Data Sciences

First Advisor

Hagop Atamian

Second Advisor

Cyril Rakovski

Third Advisor

Adrian Vajiac

Abstract

The field of drug discovery has seen remarkable advancements over the past few decades, transitioning from traditional experimental methods to highly sophisticated computational approaches. One of the pivotal techniques in this evolution is virtual screening, which utilizes molecular docking to predict the interaction between small molecules and target proteins. This method has significantly accelerated the initial stages of drug discovery by enabling the high-throughput screening of large chemical libraries. By simulating the binding affinity and stability of potential drug candidates, virtual screening has become a cornerstone in identifying promising compounds for further development.

A notable application of virtual screening was demonstrated in the search for inhibitors of the SARS-CoV-2 main protease or known as Covid M*pro, a critical enzyme for the replication of the COVID-19 virus. Researchers employed molecular docking to virtually screen vast libraries of compounds, rapidly pinpointing several candidates with high binding affinities. This approach not only expedited the discovery process but also provided valuable insights into the structural requirements for effective inhibition of Covid M*pro protein, thereby guiding subsequent experimental validation and optimization.

Moving beyond virtual screening, the advent of de novo drug design has further revolutionized the drug discovery landscape. Leveraging the power of artificial intelligence, particularly transformer-based architectures, researchers can now generate novel drug-like molecules from scratch. These models, utilizing a transformer encoder-decoder architecture, are trained on vast datasets of known compounds and their properties, enabling them to learn the intricate relationships between chemical structure and biological activity. When coupled with reinforcement learning algorithms like Monte Carlo Tree Search (MCTS), these systems can optimize the generated molecules for desired properties, such as binding affinity, specificity, and pharmacokinetics.

An exemplary study in this domain involved the design of novel inhibitors for the COVID-19 Mpro using a transformer-based de novo design framework. The researchers used a transformer encoder-decoder model to generate potential inhibitors and employed MCTS to iteratively refine these candidates. This innovative approach yielded several novel compounds with high predicted efficacy against Mpro, showcasing the potential of combining deep learning with reinforcement learning to accelerate and enhance the drug discovery process.

In conclusion, the integration of virtual screening and de novo drug design represents a paradigm shift in drug discovery. The use of molecular docking for virtual screening allows for rapid identification of potential drug candidates, while the application of transformer models and reinforcement learning in de novo design opens new avenues for the creation of innovative therapeutics. The successful identification of inhibitors for the COVID-19 M*pro protein exemplifies the transformative impact of these technologies, heralding a new era of precision and efficiency in the search for life-saving drugs.

Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Recommended Citation

D. Ang, "Advancement in in-silico drug discovery from virtual screening molecular dockings to de-novo drug design transformer-based generative AI and reinforcement learning," Ph.D. dissertation, Chapman University, Orange, CA, 2024. https://doi.org/10.36837/chapman.000611