Predicting fitness in Mycobacterium tuberculosis with transcriptional regulatory network-informed interpretable machine learning

Ethan Bustad ¹ Edson Petry ² Oliver Gu ² Braden T Griebel ^1,3 Tige R Rustad ¹ David Sherman ⁴ Jason H Yang ^2,5* Shuyi Ma ^1,3,6,7*

• ¹ Center for Global Infectious Disease Research, Seattle Children's Research Institute, Seattle, United States
• ² Center for Emerging and Re-emerging Pathogens, New Jersey Medical School, Rutgers, the State School, Newark, United States
• ³ Department of Chemical Engineering, College of Engineering, University of Washington, Seattle, Washington, United States
• ⁴ Department of Microbiology, School of Medicine, University of Washington, Seattle, Washington, United States
• ⁵ Department of Microbiology, Biochemistry & Molecular Genetics, New Jersey Medical School, Rutgers, The State University of New Jersey, Newark, New Jersey, United States
• ⁶ Department of Pediatrics, School of Medicine, University of Washington, Seattle, Washington, United States
• ⁷ Department of Global Health, School of Medicine, University of Washington, Seattle, Washington, United States

Mycobacterium tuberculosis (Mtb) is the causative agent of tuberculosis disease, the greatest source of global mortality by a bacterial pathogen. Mtb adapts and responds to diverse stresses, such as antibiotics, by inducing transcriptional stress response regulatory programs. Understanding how and when mycobacterial regulatory programs are activated could enable novel treatment strategies for potentiating the efficacy of new and existing drugs. Here we sought to define and analyze Mtb regulatory programs that modulate bacterial fitness under stress. We assembled a large Mtb RNA expression compendium and applied these to infer a comprehensive Mtb transcriptional regulatory network and compute condition-specific transcription factor activity (TFA) profiles. Using transcriptomic and functional genomics data, we trained an interpretable machine learning model that predicts Mtb fitness from TFA profiles. We demonstrated that a TFA-based model can predict Mtb growth arrest and growth resumption under hypoxia and reaeration using RNA-seq expression data alone. This model also directly delivers the transcriptional programs driving these growth phenotypes. Thus, these integrative network modeling and machine learning analyses enable the prediction of mycobacterial fitness across different environmental and genetic contexts with mechanistic detail. We envision these models can inform the future design of prognostic assays and therapeutic interventions that can cripple Mtb growth and survival to cure tuberculosis disease.