# A case study of the application of AI to early stage drug discovery

## TL;DR

This study shows that ChatGPT can help design drug candidates early in the discovery process, producing molecules with promising predicted activity.

## Contribution

Demonstrates the use of a general-purpose AI (ChatGPT) for early-stage drug discovery tasks, including molecular ideation and candidate prioritization.

## Key findings

- ChatGPT generated EGFR inhibitors with predicted IC₅₀ values of ~10–50 nM after iterative optimization.
- A single attempt produced a de novo EGFR inhibitor with a predicted IC₅₀ of 94 nM.
- A non-covalent MCL1 inhibitor candidate achieved a docking score corresponding to a 39 nM dissociation constant.

## Abstract

Artificial intelligence (AI) has emerged as a powerful tool in drug discovery, offering the potential to expedite the design of novel therapeutics. This study evaluates the effectiveness of a general-purpose conversational AI, ChatGPT (GPT-4o), in performing three distinct drug discovery tasks, assessing its ability to assist in early-stage molecular ideation and design. In the first task, ChatGPT generated molecules starting from five low-affinity EGFR inhibitors (IC₅₀ values of 10–3.16 µM), which were iteratively optimized in a QSAR model to produce compounds with predicted IC₅₀ values of ~ 10–50 nM. In the second task, de novo design of EGFR inhibitors produced a molecule with a predicted IC₅₀ of 94 nM in a single attempt. In the third task, ChatGPT generated non-covalent MCL1 inhibitors, with a top candidate achieving a docking score corresponding to a 39 nM dissociation constant. Because AI-generated molecules often face synthetic feasibility challenges, we also identified readily available analogues from a chemical vendor. These analogues were evaluated using molecular docking (AutoDock Vina) and QSAR models, with several achieving a promising activity range of 10–100 nM across the three tasks. These results demonstrate that general-purpose AI models like ChatGPT can accelerate early-stage drug discovery by assisting in molecular ideation and candidate prioritization.

The online version contains supplementary material available at 10.1038/s41598-025-32805-1.