In cell biology, the identification of new pathways via genetic screens has been a central discovery strategy. Historically, RNAi-based screens have been used to elucidate pathways in human cells. This approach has relied on two major strategies: arrayed screens, which have high specificity but require the production of each RNAi separately creating a technical bottleneck, and pooled screens, in which production is easier, but specificity and reproducibility suffer.
To overcome these limitations, we have developed a novel screening approach termed Artificial-Intelligence Microscopy Screening (AIMS). The new platform "converts" single cells into “separate” wells by applying deep learning algorithms to detect subcellular phenotypes. In brief, a genome-wide suppressor screen is performed on cells expressing dCas9 and a photo-activatable red fluorescent protein by infection with pooled guide-RNA so that every cell will express a distinct gRNA. Using live cell microscopy, individual cells exhibiting the phenotype are identified, photoactivated and sorted. To identify phenotype-to-genotype connections, the gRNA lentiviral integration site is sequenced in single cells from the selected isolated cells. We explored regulation of the transcription factor, TFEB, which initiates lysosomal and autophagy transcriptional programs in response to the cellular metabolic state. Using AIMS, we screened for factors involved in TFEB nuclear translocation. In addition to several known hits, novel TFEB regulators were also detected, demonstrating the feasibility of AIMS. Our approach is not only a novel implementation of how machine learning can be used to explore cell biology, but also a new platform enabling phenotypic-based screening at the subcellular level.