Methods and systems for predicting DNA accessibility in pan-cancer genomes

摘要

Techniques for predicting DNA accessibility are provided. A DNase-seq data file and an RNA-seq data file of a plurality of cell types are paired by assigning the DNase-seq data file to at least an RNA-seq data file within the same biotype. The neural network is configured to be trained using a batch of paired data files, wherein configuring the neural network comprises converting the convolutional layer to a first input containing DNA sequence data from the paired data files. To generate a convolution output, wherein the fully connected layer following the convolution layer includes a gene expression level derived from RNA-seq data derived from the pair of data files. Concatenating with two inputs and processing the concatenation to generate a predicted DNA accessibility output. Trained neural networks are used to predict DNA accessibility in genomic sample inputs, including RNA-seq data for novel cell types and whole genome sequencing. [Selection diagram] Fig. 1