ABSTRACT Metagenomic sequencing has revolutionized our understanding of microbial ecosystems by enabling high-resolution profiling of microbes across diverse environments. However, the resulting data are high-dimensional, sparse, and noisy, posing challenges for downstream data analysis. Machine learning (ML) has provided an arsenal of tools to extract meaningful insights from such large and complex data sets. This review surveys the existing state of ML applications in metagenomic data analysis, from traditional supervised and unsupervised learning to time-series modeling, transfer learning, and newer directions such as causal ML and generative models. We highlight certain key challenges and delve into important issues like model interpretability, emphasizing the importance of explainable AI (XAI). We also compare ML with mechanistic models, commenting on their relative advantages, disadvantages, and prospects for synergy. Finally, we preview future directions, such as the incorporation of multi-omics data, synthetic data generation, and Agentic AI systems, highlighting the increasingly prominent role that AI and ML will play in the future of microbiome science.
Sharma et al. (Tue,) studied this question.