Deep learning methods in metagenomics: a review

Abstract The ever-decreasing cost of sequencing and the growing potential applications of metagenomics have led to an unprecedented surge in data generation. One of the most prevalent applications of metagenomics is the study of microbial environments, such as the human gut. The gut microbiome plays a crucial role in human health, providing vital information for patient diagnosis and prognosis. However, analyzing metagenomic data remains challenging due to several factors, including reference catalogs, sparsity, and compositionality. Deep learning (DL) enables novel and promising approaches that complement state-of-the-art microbiome pipelines. DL-based methods can address almost all aspects of microbiome analysis, including novel pathogen detection, sequence classification, patient stratification, and disease prediction. Beyond generating predictive models, a key aspect of these methods is also their interpretability. This article reviews deep learning approaches in metagenomics, including convolutional networks (CNNs), autoencoders, and attention-based models. These methods aggregate contextualized data and pave the way for improved patient care and a better understanding of the microbiome’s key role in our health. Author summary In our study, we look at the vast world of research in metagenomics, the study of genetic material from environmental samples, spurred by the increasing affordability of sequencing technologies. Our particular focus is the human gut microbiome, an environment teeming with microscopic life forms that plays a central role in our health and well-being. However, navigating through the vast amounts of data generated is not an easy task. Traditional methods hit roadblocks due to the unique nature of metagenomic data. That’s where deep learning (DL), a today well known branch of artificial intelligence, comes in. DL-based techniques complement existing methods and open up new avenues in microbiome research. They’re capable of tackling a wide range of tasks, from identifying unknown pathogens to predicting disease based on a patient’s unique microbiome. In our article, we provide a very comprehensive review of different DL strategies for metagenomics, including convolutional networks, autoencoders, and attention-based models. We are convinced that these techniques significantly enhance the field of metagenomic analysis in its entirety, paving the way for more accurate data analysis and, ultimately, better patient care. The PRISMA augmented diagram of our review is illustrated in Fig 1 .