Machine-learning Model Makes Predictions About Network Biology

This article discusses the development of a deep-learning model called Geneformer, which is used to map gene networks and identify potential drug targets. The model is pretrained on a large, general gene-expression data set called Genecorpus-30M, which includes about 30 million gene-expression profiles of individual cells from a broad range of human tissues. The model is self-supervised and uses an attention-based approach to gain a fundamental understanding of gene-network dynamics.