Towards a machine learning approach for automated detection of well-to-well contamination in metagenomic data
Résumé
Samples subjected to metagenomic sequencing can be accidentally contaminated during wet lab steps (DNA extraction, library preparation) by DNA from an external source (e.g.: lab reagents) or from other samples processed on the same plate (well-to-well contamination). These can lead to biased results and eventually to false conclusions if not detected. Although a critical issue, well-to-well contamination remains understudied. A few tools have been developed but suffer from several limitations, such as a lack of sensitivity.
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%). Our approach does not require negative controls, works with related samples that may naturally share strains (e.g.: mother/child), discriminates contamination sources from contaminated samples and estimates contamination rates.
However, this method is time-consuming and requires human expertise to manually inspect suspect cases. We are developing 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 metagenomic experiments more robust.
Origine | Fichiers produits par l'(les) auteur(s) |
---|