Improving AI Intuition in the Discovery of New Medicines

A collaborative effort between biomedical researchers at Novartis Institutes for Biomedical Research and Microsoft Research AI4Science is breaking new ground in teaching AI systems to facilitate medicine discovery. Their study, featured in Nature Communications, harnesses chemists’ feedback to provide intuitive guidelines for an AI model.

Exploring AI-Assisted Drug Discovery with Chemist Intuition

In a pioneering endeavor, a joint team of biomedical researchers from Novartis Institutes for Biomedical Research and Microsoft Research AI4Science is venturing into AI-powered drug discovery.

 Their study, published in the journal Nature Communications, leverages insights from chemists to offer intuitive guidance to an AI model.

Challenges in Medicines Discovery

Discovering novel medicines is a complex and time-intensive process that requires collaboration among experts from diverse domains. Medical professionals and researchers must first unravel the underlying causes of a particular ailment, while chemists and researchers subsequently seek chemical compounds capable of mitigating or preventing the disease.

Both phases of this process demand significant time and effort. In this innovative project, the research team aimed to explore whether AI applications could streamline the latter stage of drug development.

Harnessing Chemist Intuition

A crucial aspect of identifying new drugs is the intuitive sense that certain chemicals may hold promise for treating specific conditions. Translating this intuition into a code has long been a challenge. However, the advent of AI applications may be changing this landscape.

To integrate AI into the drug development challenge, the researchers sought input from 45 experienced chemists responsible for discovering medicinal drugs. These experts were asked to select chemical pairs from a list of 220 teams based on their intuitive insights garnered through years of experience in the field.

The feedback gathered from the chemists was then fed into the AI system. The AI model evaluated and ranked the chemical pairs, assigning scores based on its estimates of each pair’s potential usefulness in drug development. 

Subsequently, the highest-scoring chemical pairs were passed on to an AI-based system for designing molecules from given chemicals. The results yielded by this system were deemed promising by the research team.

Validating the Approach

The researchers further assessed their system by applying it to existing drugs on the market. This evaluation revealed a noteworthy “signal” derived from the intuition data provided by chemists. 

This compelling discovery has spurred the team’s belief in additional research and exploration in this domain.

Read the original article on PHYS.

Read more: Breaking Down Breast Cancer: Targeting PTPRD with Antibodies.