Artificial intelligence powered decoding of Plasmodium falciparum sporozoite motility patterns.

Sporozoites (SPZ), the infective mosquito stage of Plasmodium falciparum malaria, are widely studied as therapeutic targets for antibodies either endogenously induced by vaccines or administered as exogenous monoclonal antibody treatment. The unicellular protozoa are highly motile and adapted to migrate through tissues, facilitating the transition from mosquito-based skin inoculation to the infection of hepatocytes in the liver. Because of this, SPZ motility is considered a prime characteristic for SPZ viability, infectivity, and therapeutic response. To evaluate the quality of parasite batches reared in malaria insectaries and monitor innovative malaria therapies, a standardized tool is needed to rapidly assess sporozoite infectivity. Previously, we designed video-based motility analysis techniques (SMOOT) that yields a multi-metric kinematic output (9 metrics). Based on the SMOOT metrics acquired for 26,682 individual SPZ tracks (80% were used for training) we have now developed an Artificial Intelligence (AI) based motility index model (unsupervised) to easily capture nuances in sporozoite quality whilst preserving data depth. The AI-model was validated formally using common AI validation methods, such as reconstruction losses, stability and robustness evaluation. Furthermore, we applied data-driven analysis, using existing monoclonal antibody SPZ inhibition ranges, showing potential for quality control for P. falciparum SPZ batches, as well as documenting antibody inhibition response.