Large language models (LLMs), like GPT-4, have shown promise in automating functional genomics research, which aims to understand the functions and interactions of genes, according to researchers at the University of California San Diego School of Medicine. Gene set enrichment, the most popular method in functional genomics, compares experimentally obtained gene sets to existing genomics databases in order to ascertain the function of the gene sets. However, traditional databases frequently do not include more intriguing and innovative biology. Scientists may be able to automate one of the most popular techniques for comprehending how genes interact to affect biology by using artificial intelligence (AI) to evaluate gene sets, which might save them many hours of labor-intensive work.
The researchers tested five different LLMs and discovered that GPT-4 performed the best, finding common functions of curated gene sets from a popular genomics database with an accuracy rate of 73%. GPT-4’s ability to assess gene sets with little delusion was demonstrated when it was asked to examine random gene sets and in 87% of situations, it refused to provide a name. Additionally, GPT-4 might include thorough explanations to support its naming scheme.
Also Read: AI is more accurate than human specialists at predicting outcomes of science research
The work emphasizes the necessity for ongoing investment in the development of LLMs and their applications in genomics and precision medicine, even if further investigation is required to fully explore the potential of LLMs in automating functional genomics. In order to assist other researchers in integrating LLMs into their functional genomics procedures, the researchers developed an online portal. In a broader sense, the results also show how AI might transform science by combining complicated data to provide novel, testable ideas in a fraction of the time.
Trey Ideker, Ph.D., a professor at UC San Diego School of Medicine and UC San Diego Jacobs School of Engineering, Dexter Pratt, Ph.D., a software architect in Ideker’s group, and Clara Hu, a PhD candidate in biomedical sciences in Ideker’s group, conducted the work, which was published in Nature Methods. The National Institutes of Health provided some funding for the study.
Source: UC San Diego Today
Journal Reference: Hu, Mengzhou, et al. “Evaluation of Large Language Models for Discovery of Gene Set Function.” Nature Methods, 2024, pp. 1-10, DOI: https://doi.org/10.1038/s41592-024-02525-x.
Last Modified:
