CGIDR Stories

Publication Q&A: Using Artificial Intelligence to Predict Drug Interactions and More Effective Treatments for Tuberculosis

November 2019 – Dr. Shuyi Ma shares insights from a recent publication in mBio with contributing authors from the Sherman Lab at the Center for Global Infectious Disease Research (CGIDR).

Transcriptomic Signatures Predict Regulators of Drug Synergy and Clinical Regimen Efficacy against Tuberculosis

Shuyi Ma, Suraj Jaipalli, Jonah Larkins-Ford, Jenny Lohmiller, Bree B. Aldridge, David R. Sherman, Sriram Chandrasekaran

Published in mBio: October 2019

Read the publication on mBio

What is your lab’s current research focus?

The Sherman lab is using experimental and computational systems biology tools to understand how the pathogen Mycobacterium tuberculosis (MTB) responds to stresses encountered during infection and to drug exposure. We are studying this pathogen in part to devise new treatment strategies for the disease that it causes: tuberculosis (TB), which is the world’s deadliest infectious disease, killing approximately 2 million people each year. The current treatment involves taking at least 4 drugs for at least 6 months, and almost all of these drugs were discovered prior to the 1980s. Our strategy is to use bacterial gene expression and the gene regulatory network as a means to dissect the molecular mechanisms that control the stress and drug response outcomes.

What is the significance of the findings of this publication?

INDIGO-MTB is a novel artificial intelligence-guided tool to inform the rational design of multidrug therapy for TB.  This tool can accurately predict interactions between drug combinations (i.e. synergy, in which the activity of the combination is greater than the sum of the individual drugs, or antagonism, in which the activity of the combination is lesser than the sum of the individuals), and also identify the genes that control these drug responses.

Today’s regimens were designed empirically, driven largely by clinical intuition, but that approach is far too slow and imprecise. The approximately 28 drugs currently in clinical use or pre-clinical development for TB can be combined into 24,000 different 3- or 4-drug regimens. Given the cost and complexity of testing each one, traditional methods to measure synergy and develop new anti-TB regimens are woefully overmatched. INDIGO-MTB addresses the huge need for efficient, data-driven approaches to prioritize new drug combinations for treating TB.

INDIGO-MTB offers several key advantages over existing tools in facilitating clinical translation. First, predictions do not require drug target information, which makes this tool compatible with designing regimens for compounds in early-phase drug development. Second, predictions are accurate for regimens featuring two or more drugs, which is critical for designing anti-TB therapies (which are made up of 4 or more drugs). Third, the input data needed to calculate predictions for new drugs are highly modular, requiring only the bacterial response for the new individual drug to predict interactions between all possible combinations with previously profiled drugs. This streamlines the effort and resources required to make new predictions.

What are the next steps for this research?

We want to improve the model’s ability to directly predict combinatorial efficacy in animals and humans for MTB, so that it can better inform the multidrug therapy design. We are also working with Dr. Rafael Hernandez (CGIDR faculty) to extend the underlying modeling approach to be able to predict efficacy for drug regimens treating other recalcitrant pathogens, including Mycobacteria absessus, a pathogen related to MTB that commonly infects cystic fibrosis patients in the U.S.

Seattle Children’s CGIDR contributing authors:

  • Shuyi Ma, PhD, research scientist
  • Jenny Lohmiller, research technician
  • David Sherman, PhD, professor