This variability of results is due to the usual statistical methods applied in compound screens, according to Dr. Genovesio, who suggests that these methods are not strictly applicable to RNAi screens (given the high variation of phenotypes that results from siRNA genome-wide knockdowns). With more than 20,000 genes in the human genome and the de facto inclusion of the numerous false-positive and false-negative results to be expected from the systematic knockdown of all gene functions within a genome, RNAi studies require multiple experiments at high resolution for meaningful conclusions to be drawn.
This translates to the need for massively high-throughput and high-content studies conducted several times over, creating anew for the biopharma field the requirement of high-throughput capacities on a whole new level. As Dr. Genovesio describes it, one must ensure “very sensitive but robust selection” with primary screens on multiple criteria, and hence, his team’s approach—high-throughput imaging.
Honing cellular microarray technology developed by David Sabatini at Whitehead Institute, Dr. Genovesio says his team has developed a platform to visually screen half a million siRNA experiments in two weeks. He stresses that the “visual” aspect of these experiments (allowing the cells themselves to be seen and phenotypes to be properly assessed) is a novel development. Similar screening by traditional methods would take up to four months and be very cost-intensive, he asserts.
Dr. Genovesio’s research team has adopted a two-pronged approach to identifying novel viral-infection targets: high-throughput siRNA cellular microarrays and a high-throughput visualization platform, with accompanying computer algorithms to parse out results from data noise.
“We’ve increased performance in two dimensions simultaneously, that is, for both imaging and the number of experiments possible. This is essential for siRNA studies. Each time we do a genome-wide screen, we conduct the same experiments as many as seven or eight times; this can mean, at the microarray level, as many as 200,000 experiments in total that need to be visualized, and their data assessed, for a single genome-wide assay.
“We have developed algorithms that can identify the experiments associated with promising drug targets through actual pictures. This is very useful for RNAi screens, because we can recover meaning more precisely: describing each experiment with lots of texture- and shape-based quantitative descriptions computed on the nuclei and cytoplasm of cells.
“We can then take these selected experiments further, and know more conclusively that, for example, in the case of infectivity studies, the cells whose phenotypes appear normal are actually representative of repressed infectivity and not just false positives. The robustness is much higher.”
Dr. Genovesio’s team at Institut Pasteur Korea is further developing its analysis platform for what he has dubbed “target deconvolution” that will essentially reverse the drug-discovery process. The approach begins with currently known effective pharmaceutical agents and “back-analyses,” through RNAi analysis and ultrahigh-content and ultrahigh-throughput visualized screening, to precisely determine networks and pathways involved in drug activity. New drug targets and their specific activity within subcellular networks are thus identified with degrees of inherent validation.