Drug candidates make up a small portion of known compounds. To find the candidates among the dead ends, extensive compound libraries are tested against a given target in high-throughput screens, with the hopes that at least one will be active. To increase the efficiency of the high-throughput screens, researchers are looking to computational resources for help. If researchers can build accurate models that can examine compounds and predict activities, they can cut the amount of compounds they would have to screen to only those with the most potential, saving a great deal of time and resources.
To this end, we developed a new pipeline. Given some previously obtained data, the pipeline will develop quantitiave structure-activity relationship (QSAR) models capable of predicting a compound’s activity. Implemented in virtual high-throughput screens, the models can single out compounds which have high predicted activity, focusing the experimental library of compounds to those that have high predicted activity, reducing the amount of compounds tested and the reagents needed to test them.
To test our pipeline, we took NCBI PubChem Bioassay AID 825 as a test case. AID 825 looks at Cathepsin-L, a receptor implicated in several disease pathways, including Ebola. Using classification and activity data from the assay, our pipeline trained our models and used them to search NCBI’s PubChem Compound database, looking for potential active compounds. From the pipeline results, we isolated a select group of compounds for which we have high confidence in our prediction. Of those compounds, 15 were commercially available and obtained for evaluation, verifying the value and accuracy of our pipeline and as new data for further model improvement.
See more of this Group/Topical: Pharmaceutical Discovery, Development and Manufacturing Forum