Back

An optimised serological machine learning model enabling targeted test-and-treat for Plasmodium vivax malaria

Smith, L.; Argyropoulos, D. C.; Bareng, A. P. N.; Lin, J.; Kiernan-Walker, N.; Lamont, M.; Abraham, A.; Lim, P.; Wu, K.; William, T.; Anstey, N.; Grigg, M. J.; Sattabongkot, J.; Lacerda, M.; Vahi, V.; Mazhari, R.; Mueller, I.; Longley, R.

2026-08-11 public and global health
10.64898/2026.08.09.26355326 medRxiv
Show abstract

The persistence of Plasmodium vivax is driven by the hidden reservoirs of infection, presenting a key obstacle to elimination. Antibodies persist after asexual infections are cleared from peripheral blood and therefore can indicate current and recent past infections. Here, we present a machine learning algorithm that classifies recent P. vivax infections using serological markers to identify likely hypnozoite carriers. Using serological measurements from year-long observational cohort studies conducted in three low-transmission settings (including negative controls, N=2,635), we selected optimal subsets of markers by balancing sero-diagnostic performance against assay complexity and scalability. We initially trained a random forest classifier and then subsequently we compared several machine learning classifiers. Tree-based methods consistently performed best, although differences were marginal. An online R Shiny application (PvSeroApp) was developed to automate data processing, quality control, and serostatus classification. This algorithm underpins the P. vivax serological testing and treatment (PvSeroTAT) strategy, enabling targeted anti-hypnozoite therapy and strengthening elimination efforts.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.