Case Studies

Resistell

How we developed machine learning algorithms for ultra-fast antibiotic susceptibility testing

Screenshot of the Resistell homepage on a tablet
Screenshot of the Resistell homepage on a laptop
Screenshot of the Resistell homepage on a mobile phone

Revolutionising diagnostics with real-time nanomotion analysis

Resistell’s Phenotech platform is redefining antibiotic susceptibility testing (AST) with a breakthrough method that detects cellular vibrations at the nanoscale. WAAT plays a key role in the platform’s success by designing and engineering the machine learning algorithms that power real-time classification of bacterial responses to antibiotics. Our contribution enables a predictive modelling layer that is core to delivering results in hours rather than days or weeks.

The brief: Create ML algorithms to power a diagnostic platform with real-time insight

WAAT was tasked with developing the machine learning backbone of Resistell’s diagnostic platform. The goal: create a robust pipeline for analysing cellular nanomotion signals and classifying antimicrobial responses across diverse bacterial strains and conditions. The challenge was to deliver a solution that could meet stringent clinical standards in speed, accuracy, and reproducibility—within the real-time constraints of a live diagnostic device.

Digital solutions: ML-powered signal interpretation for nanomotion diagnostics

  1. Custom-built ML algorithms for AST

    We engineered a classification model pipeline to translate nanomotion signals into accurate predictions of bacterial susceptibility. Our algorithms are trained to recognise subtle shifts in vibrational variance that occur when bacteria are exposed to antibiotics.

  2. Predictive models for multiple pathogens and antibiotics

    WAAT’s models have been deployed for a range of pathogens, including E. coli, K. pneumoniae, and Mycobacterium tuberculosis. For tuberculosis, our ML system achieves 97–100% sensitivity and specificity for classifying isoniazid and rifampicin resistance.

  3. Feature selection and model tuning

    Using advanced feature extraction techniques and leave-one-out cross-validation (LOOCV), our team developed high-performance models that can distinguish subtle signal behaviours in nanomotion recordings. These predictive models adapt to different bacteria-antibiotic combinations, and can be tuned for faster turnaround and new diagnostic categories.

  4. Built-in scalability and cloud integration

    We designed the ML system to work seamlessly within Resistell’s cloud-based data analysis platform. From signal acquisition to classification output, our algorithms operate in a scalable infrastructure that supports both clinical and research workflows.

Outcome

With WAAT’s machine learning integration, Resistell’s Phenotech platform transforms the diagnostic timeline for antibiotic susceptibility from days to mere hours. The technology has demonstrated over 95% predictive accuracy for bloodstream infections, and near-perfect classification performance for tuberculosis susceptibility. Our ML systems also enable real-time phage susceptibility testing, opening new frontiers for personalised treatments and AMR research.