You are here: Home » Meet » OpenTox 2011 » posterabstracts » A Weight-of-Evidence Approach to Prioritisation based on Consensus across Multiple Sources of Information

A Weight-of-Evidence Approach to Prioritisation based on Consensus across Multiple Sources of Information

Roman Affentranger and Barry Hardy, Douglas Connect, Switzerland
Glenn Myatt, Leadscope, USA
Nina Jeliazkova, IdeaConsult, Bulgaria
Matthew Clark and Jeff Wiseman, Pharmatrope, USA

We present the results of initial work carried out within the OpenToxLink Virtual Organization, applying a Weight-of-Evidence (WoE) approach based on consensus across multiple sources of information for the prediction of adverse effects of a large set of potential antimalarial compounds. The work was carried out as part of the EU FP7 project SYNERGY[1], evaluating the support of decision dashboards and event-driven collaborative research of software developed within SYNERGY.

The starting point of this project is the dataset published in May 2010 by the Tres Cantos Medicines Development Campus of GlaxoSmithKline (GSK) encompassing over 13,500 compounds obtained from screening approximately 2 million compounds in GSK’s screening library for inhibitors of proliferation of the malaria-causing parasite Plasmodium falciparum (strain 3D7) in human red blood cells (erythrocytes).[2] The 13,500 compounds in the Tres Cantos AntiMalarial Screening (TCAMS) dataset – all shown to inhibit parasite growth by more than 80 % at a 2 µM concentration – were deposited in the ChEMBL Neglected Tropical Disease archive[3] and were made publicly available. In addition to the inhibitory activity against the P. falciparum strain 3D7, the dataset also contains the compound’s activity against the multiresistant P. falciparum strain Dd2, as well as a general experimental toxicity measure of human cytotoxicity.

The goal of the project was to prioritize the 13,500 compounds in the TCAMS dataset based on a consensus across toxicity predictions obtained from various sources as well as the cytotoxicity and antimalarial data included with the TCAMS dataset. A major source of data for building predictive models was human adverse events data compiled from the US FDA’s Adverse Events Reporting System (AERS) database. We selected 19 adverse events related to the hepatobiliary tract from the AERS database, from which we formed 6 groups. Group-based predictions were obtained in two ways: 1) combining the AERS data of the adverse events belonging to a group using a consensus approach and using the combined data to train and build predictive group-based models, and 2) training and building models on each of the 19 selected adverse events and obtaining group-based predictions from a consensus across the individual predictions. These results were combined with predictions obtained from models available through OpenTox,[4] as well as with the experimental cytotoxicity data provided with the TCAMS dataset. The consensus across all these sources of evidence was obtained in a tiered, hierarchical manner. We identified a small set of compounds of high antimalarial activity in the TCAMS dataset with little or no evidence in our data for association with adverse events.

(presenting author: Roman Affentranger)

[2] Gamo F.-J. et al., Nature 2010, 465, p. 305-10
[4], Hardy B. et al., Journal of Cheminformatics 2010, 2:7

Document Actions