Towards a machine learning approach for automated detection of well-to-well contamination in metagenomics data

International audienceCombining advances in high-throughput sequencing and Big Data, metagenomic sequencing has revolutionized our vision of the microscopic world and microbiology by allowing the characterization, without prior culture, of complex microbial communities such as the human intestinal microbiota. Thus, this technology has allowed major advances in human health in recent years by showing the link between the intestinal microbiota and certain diseases such as obesity, diabetes, cancer or cardiovascular diseases.Although the laboratory protocols for processing samples are standardized and highly automated, it has been shown that during DNA extraction or sequencing, samples may be contaminated. Contamination refers to the presence of DNA that does not come from the biological sample under study. It can be due either to an external source (environmental DNA or lab reagents) or or to DNA from another sample processed at the same time. Although many solutions have been provided for the first case [4, 5], contamination between samples (well-to-well contamination) remains very little studied. However, recognition of contamination is essential : skipping this step can easily bias results and lead to false conclusions.By inspecting species abundance profiles of published cohort samples, we identified specific patterns associated with well-to-well contamination. Here, we propose an original method based on the recognition of such patterns that accurately detects contamination events even at low rates (up to 1%). This current detection procedure is based on visual inspection of potentially problematic cases. This procedure is therefore time consuming and not adapted to the massive treatment of several thousands of samples, for example in Le French Gut project. The goal of this internship is to develop a fully automated tool, based on deep learning, trained with semi-simulated sequencing data to classify contaminated samples. As preliminary results are promising, we believe this method will significantly impact the field, making metagenomics experiments more robust