ABSTRACT
There currently is renewed interest in blood clotting Factor XII as a potential target for thrombosis inhibition. Historically untargeted, there is little drug information with which to start drug candidate searches. Typical high-throughput screening can identify potential drug candidates, but is inefficient. Virtual high-throughput screening can be used to raise efficiency by focusing experimental efforts on compounds predicted to be active and is applied here to identify new Factor XIIa inhibitors. We combine principal component analysis, genetic algorithm and support vector machine to create the models used in the virtual high-throughput screening. In this work, experimental data from a PubChem Bioassay was used to train predictive models of Factor XIIa inhibition activity. The models created were then used to virtually screen the entire 72 million PubChem Compound database. Experimental validation of select candidates identified by this process resulted in a 42.9% hit-rate in the first-pass and 100% hit-rate in the second-pass, suggesting the effectiveness of the approach.
