Teaching machines to read pathogen genomes before they outsmart us

By 2050, antimicrobial resistance is projected to claim ten million lives annually, and the burden lands hardest on low- and middle-income countries. Africa CDC estimates 4.1 million deaths across the continent by then, costing up to 5% of GDP. But the number that keeps me working on this problem is smaller: three to five days. That is how long traditional susceptibility testing takes while doctors make treatment decisions half-blind.

The idea behind Requence

Requence is an end-to-end genomic intelligence platform. You hand it a pathogen DNA sequence, and a Lab-as-a-Service pipeline identifies the organism, routes it to a dedicated prediction endpoint, and returns resistance probabilities across multiple antibiotics, with the reasoning attached.

Engineering 10,000+ features from 900 genomes

We processed over 900 pathogen assemblies and engineered more than ten thousand features: amino-acid k-mers at k equal to ten, single nucleotide polymorphisms, and targeted resistance genes. Gradient boosting handles the high-dimensional, low-sample-size regime well, but only if you keep it honest with cross-validation and consensus checks across full and partial models.

Why Pearl refuses to be a black box

Our prediction model is called Pearl (my teammates named her after me). For every prediction she attaches SHAP-based explanations: which mutations drove the score, and by how much. In healthcare, a probability without a reason is close to useless. A clinician seeing +0.45 gyrA mutation impact can connect model output back to biology they already understand.

In healthcare, interpretability and speed is not a nice-to-have. It is the difference between a demo and a decision.

What is next

The second layer, Harmony, aggregates anonymized surveillance data so local patient care connects to real-time global monitoring. We are still building, refining and learning, but computational intelligence strengthening infectious disease surveillance in Africa? That is a problem worth every weekend I have given it.