Convolutional neural networks for structured omics: OmicsCNN and the OmicsConv layer

• DOI: 10.48550/arxiv.1710.05918
• Publication Date: 2017-01-01
• Authors: Giuseppe Jurman; Valerio Maggio; Diego Fioravanti; Ylenia Giarratano; Isotta Landi; Margherita France...; Claudio Agostinelli; Marco Chierici; Manlio De Domenico; Cesare Furlanello
• Keywords: 0301 basic medicine; FOS: Computer and information sciences; 570; 03 medical and health sciences; q-bio.QM; Statistics - Machine Learning; FOS: Biological sciences; Machine Learning (stat.ML); stat.ML; Quantitative Biology - Quantitative Methods; Quantitative Methods (q-bio.QM); 004

Convolutional Neural Networks (CNNs) are a popular deep learning architecture widely applied in different domains, particular classifying over images, for which the concept of convolution with filter comes naturally. Unfortunately, requirement distance (or, at least, neighbourhood function) input feature space has so far prevented its direct use on data types such as omics data. However, number metrizable, i.e., they can be endowed metric structure, enabling to adopt convolutional based framework, e.g., prediction. We propose generalized solution CNNs data, implemented through dedicated Keras layer. In particular, metagenomics derived from patristic phylogenetic tree. For transcriptomics we combine Gene Ontology semantic similarity and gene co-expression define distance; function is defined multilayer network where 3 layers by GO mutual while fourth one co-expression. As general tool, enabled OmicsConv, novel layer, obtaining OmicsCNN, framework. Here demonstrate OmicsCNN gut microbiota sequencing Inflammatory Bowel Disease (IBD) 16S first synthetic then collection 222 IBD patients.