Abstract

Neuroactive compounds are important tools for dissecting nervous system function and treating psychiatric and neurological illnesses. The major barrier to discover novel neuroactive compounds has been the lack of validated targets and the absence of an effective high-throughput in vivo screening assay. Caenorhabditis elegans provides an excellent model organism amenable to small molecule screens for probing neural and molecular basis of behaviour. Here we describe a high-throughput, behavioural-based drug screening pipeline using a unified imaging and tracking technology to examine the effects of 82 known compounds on photo-evasive response in C. elegans. To improve worm detection, we embedded a trained Convolutional Neural Network (CNN)-based classifier into the tracking system that effectively distinguishes worms from non-worm objects in recordings. We found that the dominant response to whole-body blue light illumination is an immediate increase in speed in forward movement. We showed that selective neuroactive compounds with different mechanisms of action (MOA) could alter worm photo-evasive responses in distinct ways. An XGBoost-based classification model optimized by simple majority voting correctly predicted the known MOA for 10 out of 18 test compounds based on 256 features of worm behaviour. Our preliminary results demonstrate the potential of combining high-throughput behavioural phenotyping and machine learning analytic frameworks to rapidly screen a large number of compounds and determine their MOA in the early phase of drug discovery. Additional studies need to be done to build a more accurate model to predict MOA of small molecules and to experimentally validate the molecular targets of hit compounds.

Host Labs: LMS-MRC Behavioural Genomics Research Group at Imperial College London

Biologically-Inspired Computation & Inference in the Department of Bioengineering at Imperial College London