中央研究院 Logo

The continuous decrease in sequencing costs and the growing potential applications of metagenomics have led to an unprecedented increase in data generation. One prominent application of metagenomics is the study of microbial environments, such as the human gut. As the gut microbiome plays a pivotal role in human health, its quantification paves the way for precision medicine based on metagenomic data. Indeed, machine learning is increasingly used to build predictive models of diseases, including metabolic, inflammatory, gastrointestinal, or cancerous diseases. We will provide a brief overview of the tools available today for learning such models from a microbiologist's point of view. Nevertheless, the application of these models at the patient's bedside remains challenging due to factors related to access time and the incompleteness of reference catalogs used for metagenomic quantification. Deep learning (DL) offers new and promising approaches both in terms of quantification and model performance. We will show how these approaches rethink the representation of biological knowledge in the form of "embeddings" and we will discuss some of the bioinformatics challenges of learning these powerful representations.