Andreas Pfenning-Machine learning in identifying convergent evolution associated with vocal learning скачать в хорошем качестве

Andreas Pfenning-Machine learning in identifying convergent evolution associated with vocal learning

Genome conservation is a powerful tool to annotate new genomes, prioritize trait-associated genetic variants within a population, and to link differences in traits across species to differences in selective pressures. The vast majority of computation methods to infer conservation rely on the alignment of individual nucleotide sequences. While these approaches work well for many protein sequences and highly conserved non-coding regions, they fail at the vast majority of enhancers. These distal regulatory elements are often conserved in their cell type- and tissue-specific function, even when nucleotide conservation is low. To overcome this limitation, we developed the TACIT (Tissue Aware Conservation Inference Toolkit) approach, in which machine learning models learn the regulatory code connecting genome sequence to tissue-specific open chromatin, allowing us to accurately predict cases where differences in genotype are associated with differences in tissue-specific open chromatin at enhancer regions. We trained convolutional neural network models on brain open chromatin from the Pfenning laboratory and publicly available datasets to predict open chromatin across orthologous regions of 222 mammals. Within the human population, we demonstrate that conserved neural cell type-specific open chromatin provides a substantial boost to prioritizing and interpreting disease-associated genetic variants. Across mammals, we identify a number of cases where predicted brain open chromatin is associated with vocal learning, the ability to modify vocal production based on auditory input. These vocal learning-associated non-coding regions tend to be found near transcription factors, including several that have been implicated in speech production deficits in the human population.