PERFect: PERmutation Filtering test for microbiome data

• DOI: 10.1093/biostatistics/kxy020
• Publication Date: 2018-05-30T03:37:27Z
• Authors: Ekaterina Smirnova; Snehalata Huzurbazar; Farhad Jafari
• Keywords: Replicate

The human microbiota composition is associated with a number of diseases including obesity, inflammatory bowel disease, and bacterial vaginosis. Thus, microbiome research has the potential to reshape clinical therapeutic approaches. However, raw count data require careful pre-processing steps that take into account both sparsity counts large taxa are being measured. Filtering defined as removing present in small samples have where they observed. Despite progress quality filtering approaches, there no consensus on standards assessment. This can adversely affect downstream analyses reproducibility results across platforms software. We introduce PERFect, novel permutation approach designed address two unsolved problems processing: (i) define quantify loss due by implementing thresholds (ii) evaluate test for provide measure excessive filtering. Methods assessed three "mock experiment" sets, true compositions known, applied publicly available real sets. method correctly removes contaminant "mock" quantifies visualizes corresponding loss, providing uniform data-driven criteria In PERFect tends remove more than existing approaches; this likely happens because based an explicit function, uses statistically principled testing, takes correlation between taxa. software freely at https://github.com/katiasmirn/PERFect.