Multiplexed Experiment Design in High-Throughput Screening

An early step in the drug discovery involves screening through numerous chemical compounds to find molecules that have a specified biological or biochemical effect. This screening process is costly, thus any modifications to make the screen more efficient would reduce the total cost of drug discovery. Current high-throughput screening methods test each compound from a library of compounds individually on a biological assay. This implies that 100, 000 assays are required to tests 100, 000 compounds, even though only a small number of the compounds are active. In this paper we outline an alternative experimental design similar to group testing that allows a compound library to be tested in fewer experiments. Instead of the usual construction methods for designing such experiments we use a flexible machine learning approach for experiment design. Our method is called BEST — Bayesian Entropy Search Technique. BEST uses a Bayesian posterior calculation and an information-theoretic metric, entropy, to score an experimental design and then optimizes the design through a modified genetic algorithm. Using BEST , we found experiments that can detect between 0 and 2 drugs in a compound library of 20 compounds with 100% accuracy using ≥ 11 assays with ≤ 5 compounds per assay. This is a 45% reduction compared to a non-multiplexed screening approach. Using BEST , we demonstrate that optimal experiment designs use significantly fewer assay measurements and can still be constrained on the number of compounds that can be mixed in one experiment. The immediate application of these experimental designs include the creation of pre-mixed libraries of compounds based on our best design that can be tested on several targets and the results decoded via the scheme described in this paper.