Machine Learning Model Predicts Network Biology Patterns

Scientists have developed a machine-learning model called Geneformer that predicts gene interactions and cell states using a large, general gene-expression data set. This helps map gene networks, which are vital in understanding biological and disease processes and identifying medical targets. Gene-network mapping needs significant gene-expression data, but data is often limited. The model is designed using transfer learning and a deep-learning model that gains a fundamental understanding of gene-network dynamics to forecast gene interactions. Pretraining allowed Geneformer to encode the hierarchy of gene networks, which boosted its predictive accuracy compared with standard alternative approaches. Geneformer’s context-awareness enables it to make disease-specific predictions in distinctive situations. As more publicly available gene-expression data expands, more precise predictions can be made. The model may also be used to study diseases in which different cell types are affected, and progressive diseases where therapeutic targets might differ depending upon disease stage.