Decoding yeast transcriptional regulation via a data- and mechanism-driven distributed large-scale network model

Abstract

The complex transcriptional regulatory relationships among genes influence gene expression levels and play crucial roles in determining cellular phenotypes. Here, we present a novel distributed large-scale transcriptional regulatory neural network model (DLTRNM) by integrating prior knowledge into the reconstruction of pre-trained machine learning models followed with fine-tuning. Taken S. cerevisiae as the example, the transcriptional regulatory relationships, carefully compiled and documented, are utilized to define interactions between TFs and their TGs within DLTRNM. Subsequently, DLTRNM is pre-trained on pan-transcriptomic data and fine-tuned with time-series data, enabling it to accurately predict dynamic regulatory correlations between TFs and TGs, such as simulating TG responses by tuning the expression of TFs. In addition, DLTRNM can assist in identifying potential key TFs for a subset of genes, thus simplifying the complex and interrelated transcriptional regulatory networks (TRN). Based on the key TFs, it can also refine previously reported transcriptional regulatory subnetworks and highlight the core regulatory networks. DLTRNM offers a powerful tool for studying transcriptional regulation with reduced computational demands and enhanced interpretability. Thus, this study represents a significant step forward in systems biology for understanding the complicated transcriptional regulation within cells.