A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action

Abstract

Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated ''white-box'' biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genomescale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy.

Chromosome-free bacterial cells are safe and programmable platforms for synthetic biology

“…It is also found that the TCA cycle was largely shut down. This enhanced carbon metabolic response to energy starvation was also observed in other studies where the cell experienced DNA damage and launched a stress response (44). With the deletion of the chromosome, we reduced the functional space of the cell's metabolic network (45), which resulted in a less complex interactome (the whole set of molecular interactions) and streamlining of the cellular processes in SimCells.…”

The Biophysics of Regenerative Repair Suggests New Perspectives on Biological Causation

“…It is likely that machine learning (AI) will be a critical part of the future of this field—not as an endpoint (via black box prediction), but as the first step of a profound understanding, by helping to identify richer kinds of causes as new testable hypotheses. The development of artificial intelligence platforms—the next generation of a bioinformatics of shape—could help identify potent interventions that enable control of form, and thus reveal the causal structure of complex biological systems that can then be investigated mechanistically . Conversely, a more nuanced understanding of causation is itself critical for the development of novel machine learning strategies that extract actionable intelligence from the ever‐increasing deluge of data …”

Biological machine learning combined with bacterial population genomics reveals common and rare allelic variants of genes to cause disease

“…XGboost can identify genetic variants in human GWAS as demonstrated in a Finnish study that integrated complex nonlinear interactions of SNPs 10 . The ability to interrogate the predictive features enables whiteboxing the parameters, which is emerging as a tool for deriving mechanistic function in biology 11 . XGboost implements adaptive optimization within the functional space by iteration of the weak learners into strong learners represented by decision trees where each new decision tree is generated by factoring the residual generated from the difference from observed to the predicted feature (Figure 2; Supplemental Table 1).…”