Kuraku, S. — Page 3 | Search bioRxiv

bioRxiv2018

Evolution of Nodal and Nodal-related genes and the putative composition of the heterodimer that triggers the Nodal pathway in vertebrates

Nodal is a signaling molecule that belongs to the transforming growth factor-beta superfamily that plays key roles during the early stages of development of animals. Nodal forms an heterodimer with a GDF1/3 protein to activate the Nodal pathway. Vertebrates have a paralog of nodal in their genomes labeled Nodal related, but the evolutionary history of these genes is a matter of debate, mainly because of variable numbers of genes in the vertebrate genomes sequenced so far. Thus, the goal of this study was to investigate the evolutionary history of the Nodal and Nodal-related genes with an emphasis in tracking changes in the number of genes among vertebrates. Our results show the presence of two gene lineages (Nodal and Nodal-related) that can be traced back to the ancestor of jawed vertebrates. These lineages have undergone processes of differential retention and lineage-specific expansions. Our results imply that Nodal and Nodal-related duplicated at the latest in the ancestor of gnathostomes, and they still retain a significant level of functional redundancy. By comparing the evolution of the Nodal/Nodal-related with GDF1/3 gene family, it is possible to infer that there are at least four types of heterodimers that can trigger the Nodal pathway among vertebrates.

evolutionary biology↗

bioRxiv2018

A regulatory-sequence classifier with a neural network for genomic information processing

Genotype-phenotype mapping is one of the fundamental challenges in biology. The difficulties stem in part from the large amount of sequence information and the puzzling genomic code, particularly of non-protein-coding regions such as gene regulatory sequences. However, recently deep learning-based methods were shown to have the ability to decipher the gene regulatory code of genomes. Still, prediction accuracy needs improvement. Here, we report the design of convolution layers that efficiently process genomic sequence information and developed a software, DeepGMAP, to train and compare different deep learning-based models (https://github.com/koonimaru/DeepGMAP). First, we demonstrate that our convolution layers, termed forward- and reverse-sequence scan (FRSS) layers, enhance the power to predict gene regulatory sequences. Second, we assessed previous studies and identified problems associated with data structures that caused overfitting. Finally, we introduce several visualization methods that provide insights into the syntax of gene regulatory sequences.

bioinformatics↗