2

Convolutional neural networks on biological sequences are known to learn probabilistic motifs associated with the target phenotype. To go beyond this informal interpretation, we provide a statistical test quantifying the motif-phenotype association. This requires to solve a post-selection inference problem, where the selection involves the infinite set of possible motifs.

We analyze convolutional and recurrent neural networks for biological sequences in the framework of positive definite kernels. In addition to providing a new interpretation on the genomic features underlying these networks, this analysis provides a version that is better regularized and performs better when few data is available. We also extended this work to graph-structured data.

We describe the variation in bacterial genomes through their content in short sequences (k-mers). This avoids the bias usually caused by focusing on pre-defined SNPs or gene lists. We use this representation to pinpoint the genomic variation associated to antimicrobial resitances, and rely on a so-called de Bruijn graph to help make sense of the identified k-mers in terms of SNPs or mobile genetic elements.