StanDep: capturing transcriptomic variability improves context-specific metabolic models

Diverse algorithms can integrate transcriptomics with genome-scale metabolic models (GEMs) to build context-specific metabolic models. These algorithms rely on preprocessing - identifying a list of high confidence (core) reactions from transcriptomics. Studies have shown parameters related to preprocessing, such as thresholding of expression profiles, can significantly change model content. Importantly, current thresholding approaches are burdened with setting singular arbitrary thresholds for all the genes; thus, resulting in removal of enzymes needed in small amounts and even many housekeeping genes. Here, we describe StanDep, a novel heuristic method for preprocessing transcriptomics data prior to integration with metabolic models. StanDep clusters enzymes based on their expression pattern across different contexts and determines thresholds for each cluster using data-dependent statistics, specifically standard deviation and mean. Hundreds of models for the NCI-60 cancer cell lines, human tissues, and C. elegans cell types were built using StanDep. These models were able to capture higher number of housekeeping genes and improved precision in predicting gene essentiality (CRISPR and RNAi) compared to models built using more established approaches. Our study also provides novel implications for understanding how cells may be dealing with context-specific and ubiquitous functions.