Digital Phenotyping of Tuberculosis and HIV Infection
In this pilot collaboration study, we demonstrate the power of QES in detecting TB phenotypes with <5ul of plasma, with no sample prep, no reagents, and only one pipetting step.
We also demonstrate how a digital twin of the sample can reveal novel insights (HIV phenotyping) over the air, without consuming additional samples.
Conference: AMP 2022
Using the Digital Phenotype of Disease in Human Plasma for Simplified and Accurate Prediction of Tuberculosis and HIV Infection
Introduction: Approximately 1.5 million died from tuberculosis (TB) disease in 2020. Nucleic acid amplification tests (NAAT) have greatly improved the detection of the Mycobacterium tuberculosis and HIV comorbidity. However, the frequently challenging sputum collection for NAAT creates a risk of infection, restricting testing to facilities that are far removed from the patients. NAAT workflows also require expensive instrumentation and consumables suited to centralized and better-resourced labs. To address these limitations, we developed a new, label- and probe-free electrochemical sensing modality, Quantum Electrochemical Spectroscopy (QES), for the rapid detection of TB and HIV infection from 2 µl of patient plasma in 40 minutes. Methods: A total of 2 µl aliquots of human plasma were pipetted into a consumable sensor coupled to a Probius QES instrument that performs a voltage scan of the sample at a nanoscale electrochemical interface and measures the resulting charge transfer current, which contains discrete, vibration-correlated signatures of species in the sample. We collected QES signatures from a cohort of 30 training and 10 blinded samples from South African symptomatic patients. The training samples were equally portioned between rule-out and TB-positive samples. Blood and sputum samples collected from pulmonary TB patients and rule-out individuals were assigned TB phenotypes after analysis of their sputum by the culture and or GeneXpert System. A separated subset of the training samples (~10) was used for validation of a trained k-nearest neighbor’s classifier. Once validated, the classifier was used to predict the TB disease phenotype on the 10 blind samples. In addition, without re-assaying the samples with new reagents, we developed an HIV classifier using the same training and validation protocols, by adding available information on patient HIV burden to the training data and re-training in silico. Results: Our method demonstrated 90% overall accuracy on the 10 validation and 10 blinded samples for the TB phenotype. When repurposing the data