AI-enhanced LVSim2.0 technology for rapid phenotypic susceptibility testing of new and experimental TB drugs
openNIAID - National Institute of Allergy and Infectious Diseases
PROJECT SUMMARY
Tuberculosis (TB) and multidrug-resistant TB (MDR-TB) remain major global health threats, with over 10 million
cases and 1.5 million deaths annually. MDR-TB presents a significant barrier to TB control, with nearly 20% of
affected patients dying within a year of treatment initiation. Accurate and timely drug susceptibility testing (DST)
is essential to guide therapy and prevent ineffective treatment, yet current DST methods are insufficient.
Molecular resistance assays, though rapid, are limited by incomplete knowledge of TB resistance mechanisms,
rendering them ineffective for detecting resistance to new, repurposed, or experimental pre-clinical TB drugs. In
contrast, phenotypic DST remains the gold standard, as it directly measures bacterial growth in the presence of
antibiotics. However, traditional phenotypic methods are slow, culture-based, and labor-intensive, delaying
treatment decisions by weeks. The long-term goal is to advance a universal, simple, rapid phenotypic DST
technology that can integrate processed raw samples and determine TB susceptibility or resistance to any new,
repurposed, or clinical trial drug, thereby ensuring that the appropriate choice of drug treatment is determined
and executed as early as possible in the time course of MDR-TB disease. To address this critical gap, we
propose to develop Large Volume Scattering Imaging (LVSim) and rapid machine-learning-based TB phenotypic
DST (LVSim-TBDST), a high-throughput, universal, label-free, and rapid phenotypic DST technology. LVSim-
TBDST can determine TB drug susceptibility independent of genetic markers and detect heteroresistance at the
therapeutic failure threshold. Our central hypothesis is that LVSim-TBDST can rapidly and accurately assess TB
drug susceptibility using scattering-based optical imaging and advanced deep learning to analyze bacterial
growth dynamics. This approach eliminates the need for molecular labels, genetic markers, or biochemical
staining while significantly reducing time-to-result for phenotypic DST. By applying advanced imaging techniques
and data-driven analysis, LVSim-TBDST has the potential to revolutionize universal TB drug susceptibility
testing, particularly for new and repurposed drugs lacking molecular resistance assays. The project will 1)
engineer next-generation LVSim2.0 optical sensing technology for microplate-based, high-throughput, label-free
rapid phenotypic DST, 2) establish LVSim2.0 technology and develop the AI-driven image processing algorithms
for rapid TB pDST with new TB drugs and for detecting 1% heteroresistant TB populations, and 3) develop a
workflow for direct from mycobacterial clinical sample LVSim2.0 pDST TB testing. The project will be carried out
by a productive multidisciplinary scientific team with over a decade of collaborative research experience and
extensive expertise in 1) biosensors and engineering, 2) tuberculosis, clinical microbiology, and diagnostics, and
3) bioanalytical instrument development and production. The results will have a positive impact immediately
because this technology universally performs TB phenotypic DST to promptly inform clinical decisions for
effective treatment of MDR-TB patients.
Up to $2.4M
health research