A machine learning approach to define antimalarial drug action from heterogeneous cell-based screens

Abstract

Drug resistance threatens the effective prevention and treatment of an ever-increasing range of human infections. This highlights an urgent need for new and improved drugs with novel mechanisms of action to avoid cross-resistance. Current cell-based drug screens are, however, restricted to binary live/dead readouts with no provision for mechanism of action prediction. Machine learning methods are increasingly being used to improve information extraction from imaging data. Such methods, however, work poorly with heterogeneous cellular phenotypes and generally require time-consuming human-led training. We have developed a semi-supervised machine learning approach, combining human-and machine-labelled training data from mixed human malaria parasite cultures. Designed for high-throughput and high-resolution screening, our semisupervised approach is robust to natural parasite morphological heterogeneity and correctly orders parasite developmental stages. Our approach also reproducibly detects and clusters drug-induced morphological outliers by mechanism of action, demonstrating the potential power of machine learning for accelerating cell-based drug discovery.One Sentence Summary: A machine learning approach to classifying normal and aberrant cell morphology from plate-based imaging of mixed malaria parasite cultures, facilitating clustering of drugs by mechanism of action. Ashdown et alPage 3Cell-based screens have significantly advanced our ability to find new drugs (1). However, most screens are unable to predict the mechanism of action (MoA) of identified hits, necessitating years of follow up post-discovery. In addition, even the most complex screens frequently discover hits against cellular processes that are already targeted (2). Limitations in finding new targets is becoming especially important in the face of rising antimicrobial resistance across bacterial and parasitic infections. This rise in resistance is driving increasing demand for screens that can intuitively find new antimicrobials with novel MoAs. Demand for innovation in drug discovery is exemplified in efforts targeting Plasmodium falciparum, the parasite that causes malaria. Malaria disease continues to be a leading cause of childhood mortality, killing nearly half a million children each year (3). Drug resistance has emerged to every major antimalarial to date including rapidly emerging resistance to frontline artemisinin-based combination therapies (4). Whilst there is a healthy pipeline of developmental antimalarials, many target common processes (5) and may therefore fail quickly because of prevalent crossresistance. Thus, solutions are urgently sought for the rapid identification of new drugs that have a novel MoA at the time of discovery. Parasite cell morphology within the human red blood cell (RBC) contains inherent MoA predictive capacity. Intracellular parasite morphology can distinguish broad stages along the developmental continuum of the asexual parasite (responsible for all disease pathology). This developmental continuum includes e...

4D live-cell imaging of microgametogenesis in the human malaria parasite Plasmodium falciparum

“…Ultrastructure expansion microscopy was recently shown to advance traditional fluorescence microscopy approaches to fixed parasite imaging, allowing close observation of asexual blood stage, microgametocyte and ookinete cytoskeletal development 22 . A recent study reported the application of semi-supervised machine learning to define asexual parasite development in a high-throughput imaging format, using fixed parasites 24 . The study proved to be a powerful tool in detecting parasites with morphological perturbations when treated with known antimalarials 24 .…”

Automated detection and staging of malaria parasites from cytological smears using convolutional neural networks

“…However, the ground truth could be further refined by including more annotators and allowing annotators to select from a wider range of stages. Moreover, our pipeline could be supplemented by asking annotators to order cells as part of the labelling process as this has been shown to improve consensus among annotators in fluorescent imaging data of P. falciparum (26). It should be noted that there is as yet no method of validating this ground truth; ultimately, it is defined by the experts who evaluate the slides, and this phenomenon has led groups to correct standard datasets (41).…”

“…Only a few studies have also combined parasite detection with the classification of the different stages of the intraerythrocytic development cycle (IDC). Furthermore, treating parasite development as a classification problem disregards information on progression within and between the individual stages; progression through the IDC is a continuous process and experts disagree on the boundaries between the different life stages (10, 21, 25, 26). Automation has the potential to save time and add to number of RBCs sampled for both diagnosis and lab usage, however, its usage in diagnosis would require confidence in the analysis process; if results are accessible for review post-analysis by a microscopist, such a system is more likely to be implemented as a robust decision support tool.…”

Identification and characterisation of a phospholipid scramblase in the malaria parasite Plasmodium falciparum

“…6 A). To generate the donor plasmid pKW003_1022700, primers 1022700 5 F and R were used to amplify the 572 bp 5’ homology flank and primer pair 1022700 3 F and R was used to amplify the 673 bp 3’ homology flank (KOD Hot Start DNA Polymerase, Merck Millipore) which were cloned on either side of the sfGFP expression cassette in pkiwi003 [ 35 ] using appropriate restriction sites (Table S1). The pDC2-cam-coCas9-U6.2-hDHFR vector [ 36 ] was used to clone in either of the guide RNAs 951969, 952497 and 952530, which were identified by using the web tool CHOPCHOPv2 ( http://chopchop.cbu.uib.no [ 37 ]).…”