Sparse identification of bacterial transcriptional regulation

In bacteria, the relationship between a gene's regulatory structure and its expression is well-understood for specific gene circuits. However, a comprehensive, genome-scale view of this relationship is lacking. A recent collaborative project from our lab used bacterial single-cell RNA sequencing (scRNA-seq) and single molecule fluorescence in situ hybridization (smFISH) to measure the genome-wide cell cycle pattern of Escherichia coli (E. coli) transcription (PMID: 37034646). While the transcriptional activities of many genes align with a null model where transcription rates simply mirror gene dosage (PMID: 26669443), some genes exhibit variations, such as shifts in timing or amplitude of expression. As a follow-up, we aim to use machine learning-aided methods to construct ordinary differential equation (ODE) models of transcription rates, which reveal regulatory mechanisms beyond the null model. After evaluating various algorithms, we have focused on the sparse identification of nonlinear dynamics (SINDy) algorithm (PMID: 27035946) because of its superior computational efficiency. We tested SINDy with simulated mRNA number data from ODE models that represent known mechanisms, such as transcriptional activation and repression, thus validating SINDy's capability to recognize basic biological patterns. Our current effort is to apply SINDy to genome-wide E. coli transcription data from our lab’s collaborative project. We will compare SINDy-inferred models with existing databases such as EcoCyc and RegulonDB to ascertain consistency with documented regulatory mechanisms and evaluate their predictive power for unidentified mechanisms. Our attempt will serve as a starting point for learning more complicated transcriptional regulation models using data-driven methods.