Mobile Menu

Defining antimalarial drug action using machine learning

A team of researchers have developed a semi-supervised machine learning approach to define antimalarial drug action from heterogenous cell-based screens.

Antimicrobial resistance

The use of cell-based screens has substantially advanced the drug discovery process. Nonetheless, most screens are unable to actually predict the mechanism of action (MoA) of identified hits. The rise of antimicrobial resistance is an ongoing concern, with the limitations in finding new targets becoming increasingly important. The rise in resistance has driven the demand for screens that can intuitively find new antimicrobials with novel MoAs.

A prominent example can be seen in the efforts on targeting Plasmodium falciparum, the parasite that causes malaria. Malaria continues to be the leading cause of childhood mortality. Simultaneously, drug resistance to nearly every major antimalarial has emerged. Therefore, solutions for the rapid identification of new drugs that have novel MoAs are urgently needed. Machine learning methods are being increasingly used to improve information extraction from imaging data, particularly deep neural networks (DNNs). Therefore, these methods could provide a potential solution.

Machine-learning model

In this study, published in Science Advances, researchers created a semi-supervised model that discriminates between diverse morphologies across the asexual life cycle continuum of P. falciparum. The team combined human- and machine-labelled training data from mixed human malaria parasite cultures.

The team found that this semi-supervised model could define unperturbed parasite development with finer information granularity than human labelling alone. Additionally, it was able to identify effective drugs and cluster them according to MoA, based on life cycle stage and morphological outliers. The combining of life cycle and morphology embeddings allowed the DNN to group compounds based on their MoA without directly training the model on these morphological outliers.

This model approach addresses the combined need of high-throughput cell-based drug discovery that can rapidly find new hits and predict MoA at the time of identification. Application of ML-driven screens could enable the rapid, large-scale screening and identification of drugs along with determination of predicted MoA.

Image credit: By jcomp – www.freepik.com

Share this article