Isaac Mujuni

Beyond Traditional synthesis, using transformer models for molecule generation

Started Aug. 17, 2023

Abstract or project description

The generation of novel molecules is an important step in drug discovery and the creation of new therapeutic agents. Traditional approaches to molecular design often require vast time and resources, relying on iterative processes of synthesis and testing to identify candidates with therapeutic potential. However, recent advancements in artificial intelligence — particularly large language models (LLMs) — present an opportunity to revolutionize this domain. LLMs — with their capacity to understand, generate, and manipulate text-based information — can be repurposed to recognize patterns and relationships in molecular structures and bioactivity data. When trained on extensive chemical databases, LLMs can predict potential drug-like molecules, recommend modifications to existing ones, or even generate entirely novel molecular structures; this not only expedites the molecule discovery phase but also reduces costs and resources. Furthermore, by integrating LLMs with other computational methods like molecular docking and dynamics simulations, a more holistic approach to drug design can be realized. This project will entail leveraging the transformative potential of LLMs with an eye towards laying the foundations for a healthcare startup by either:

a) developing an executive summary (downside scenario);

b) clarifying product-market fit (base case); or

c) preparing an investor-ready pitch deck (upside scenario)

Understanding differences between natural and artificial "cognition"

Started Feb. 15, 2023

Abstract or project description

We are still looking for a dataset that will give definition to this project