Mobile Menu

Bayesian Machine Learning Strategy Identifies Anti-Schistosomal Small Molecules

Recent research has identified anti-schistosomal small molecules based on a Bayesian machine learning strategy for drug discovery.


Schistosomiasis is a chronic infectious disease associated with poverty and affects around 200 million people worldwide. This disease is caused by the Schistosoma species of flatworm, which lives in the blood vascular and produces eggs that cause a variety of pathologies. Effects of Schistosomiasis are painful and lifelong, and treatment relies solely on praziquantel (PZQ). Although this drug is affordable and reasonably effective at reducing morbidity, as the only treatment available, there are concerns surrounding decreased efficacy and resistance. Computational techniques are an attractive drug discovery technique given the lack of financial investment into this disease.

The Bayesian machine learning technique

Drug discovery for anti-schistosomal compounds employs in vitro whole organism screens against two developmental stages of Schistosoma; post-infective larvae (somules) and adults. Using data from existing literature, the researchers in this study generated two rule books, with an associated scoring system to normalize nearly 4,000 phenotypic screening data points to enable machine learning. Using software called Assay Central, the data was used by the researchers to generate eight Bayesian machine learning models according to the parasite’s developmental stage and experimental time points. Researchers currently use the Bayesian machine learning method to predict active compounds for other poverty associated diseases, such as Chagas disease, Ebola and Tuberculosis.

The rule books generated in this study produced a sliding scale of scores, from 0 (no activity) to 4 (most activity), where potent compounds that act quickly at low concentrations get a higher score than those that take more time or require a higher concentration. Rule book scores were used with the Assay Central technology to predict compounds for in vitro screening against Schistosoma.

The models helped predict 56 active and nonactive compounds from commercial compound libraries for testing. When screened against Schistosoma in vitro, the prediction accuracy for the compounds was 61% and 56%, with a hit rate of 48% and 34% for somules and adults, respectively. This far exceeds the typical 1-2% hit rate for traditional high throughput screens. This increased hit rate is an exciting finding, given that schistosome research requires small animal hosts to propagate the parasite and limited numbers of parasites can be recovered per host.

The rule books generated in this study helped to improve existing machine learning strategies and facilitated the first step towards a unified database of anti-schistosomal activity. Overall, the Bayesian method is faster at generating models compared to other algorithms, such as deep learning, and can be implemented quickly on desktop computers, which is a major advantage in the limited drug discovery research environment for diseases of poverty.

This research has outlined a machine learning model that can be used to investigate therapies for neglected diseases of poverty, such as Chagas disease and Trichomoniasis.

Image credit: FreePik

Share this article