Automated image-based phenotypic screening for high-throughput drug discovery

At the state-of-the-art in drug discovery, one of the key challenges is to develop high-throughput screening (HTS) techniques that can measure changes as a continuum of complex phenotypes induced in a target pathogen. Such measurements are crucial in developing therapeutics against diseases like schistosomiasis, trypanosomiasis, and leishmaniasis, which impact millions worldwide. These diseases are caused by parasites that can manifest a variety of phenotypes at any given point in time in response to drugs. Consequently, a single end-point measurement of ´live or death´ (e.g., ED₅₀ value) commonly used for lead identification is over-simplistic. In our method to address this problem, the parasites are tracked during the entire course of (video) recorded observations and changes in their appearance-based and behavioral characteristics quantified using geometric, texture-based, color-based, and motion-based descriptors. Subsequently, within the on-line setting, machine learning techniques are used classify the exhibited phenotypes into well defined categories. Important advancements introduced as a consequence of the proposed approach include: (1) ability to assess the interactions between putative drugs and parasites in terms of multiple appearance and behavior-based phenotypes, (2) automatic classification and quantification of pathogen phenotypes. Experimental data from lead identification studies against the disease Schistosomiasis validate the proposed methodology.