A joint research team from KAIST and the University of California San Diego (UCSD) has proposed an Artificial Intelligence (AI)-driven strategy designed to accelerate the discovery of microbial gene functions. This approach addresses the ongoing challenge of identifying the roles of numerous genes in microbial genomes despite extensive whole-genome sequencing data. The proposed strategy integrates computational and experimental biology, utilizing advanced AI models and an "Active Learning" framework to enhance efficiency and accuracy in gene function identification.
Research Proposal Overview
Distinguished Professor Sang Yup Lee of KAIST and Professor Bernhard Palsson of UCSD co-authored a comprehensive review paper detailing this strategy. The paper, titled "Approaches for accelerating microbial gene function discovery using artificial intelligence," was published in Nature Microbiology on January 7th. The study systematically analyzes and organizes current AI-based research methods aimed at improving the speed of gene function discovery.
Challenges in Gene Function Discovery
For two decades, whole-genome sequencing has been a routine practice, yet the functions of a significant portion of genes in microbial genomes remain unknown. Traditional experimental methods, including gene deletion, expression profile analysis, and in vitro assays, are often time-consuming and costly. These methods also encounter limitations related to large-scale experimentation, complex biological interactions, and potential discrepancies between laboratory and in vivo results.
AI-Driven Solutions
The research team advocates for an AI-driven approach that integrates computational and experimental biology. Their paper reviews various computational methods, ranging from sequence similarity analysis to deep learning-based AI models. Key AI technologies highlighted include:
- 3D Protein Structure Prediction: Tools such as AlphaFold and RoseTTAFold are noted for their potential to provide insights into the mechanisms of gene function, extending beyond mere function estimation.
- Generative AI: This technology is being utilized to design proteins with specific desired functions.
The team presented application cases and future research directions for integrating gene sequence analysis, protein structure prediction, and metagenomic analyses, with a focus on transcription factors and enzymes.
Active Learning Framework
To overcome potential biases in traditional gene discovery, the researchers advocate for an "Active Learning" framework. In this method, an AI model identifies predictions with high uncertainty and suggests specific experiments to resolve these uncertainties. The results from these experiments are then fed back into the model, iteratively improving its accuracy and efficiently validating critical gene functions.
The implementation of this framework requires integration with automated experimental platforms, shared research infrastructures like biofoundries, and the sharing of "failed data" as a learning asset.
Future Directions and Challenges
Dr. Gi Bae Kim of KAIST, a co-author, highlighted that developing "Explainable AI" models, which can provide biological justifications for their results, remains a critical challenge.
Professor Sang Yup Lee emphasized that the key to advancing gene function discovery involves combining a systematic, AI-guided experimental framework with automated research infrastructure, all under human oversight. He stressed the importance of establishing a research ecosystem where prediction and validation are repeatedly linked.
Support and Publication
The research received support from the National Research Foundation and the Korean Ministry of Science and ICT.