TriCC: Self-Supervised Contrastive Learning for Image-Based Antiparasitic Drug Discovery

The fight against parasitic worm infections requires the investigation of a vast array of potential drug candidates. One important tool in this screening is microscopic image classification. These images are acquired from model organisms such as C. elegans exposed to drug candidates. We report a dataset and a novel self-supervised method of clustering and classification to identify natural product extracts that cause Abnormal phenotypes in C. elegans. Our method, Triple Cluster Classification (TriCC), has the ability to identify patterns in C. elegans that accurately map to 27 observed phenotype combinations with a weighted average area under ROC curve of 0.8. In binary classification, to identify drugs with any impact, the area under the ROC curve is 0.89.