HCS/HCA Data and Informatics Special Interest Group (SIG)

Topic I: High-Content Imaging Probing the Real-Time Plasticity of EMT Phenotypes Driving Malignant Cancer

Discussion Leaders: Hector Esquer, MS (PhD Candidate) & Daniel V. LaBarbera, PhD, Associate Professor, Department of Pharmaceutical Sciences, The Skaggs School of Pharmacy and Pharmaceutical Sciences, University of Colorado Anschutz Medical Campus.

Epithelial-mesenchymal transition (EMT) is a driving force in cancer by promoting tumor initiation, growth, and metastasis as well as multidrug resistance to conventional cancer therapies. A major problem in understanding EMT biology and developing therapeutic intervention is that EMT is a dynamic process that can undergo a mesenchymal-epithelial transition (MET) and partial transitions between both phenotypes resulting in a broad spectrum of tumorigenic phenotypes. To address this problem, we have developed real-time fluorescent promoter driven dual reporter assays for vimentin (VimPro) and E-cadherin (E-CadPro), putative markers for the mesenchymal and epithelial phenotypes, respectively. Stable dual reporter cell lines generated in colon (SW620), breast (MDA-MB-231), and lung (A549) cancer models have revealed three distinct EMT phenotypes that display different metabolic profiles and tumorigenic properties in cell and animal models. Using high content imaging and machine learning, we show long-term cultures of these cell lines recapitulate the epithelial and mesenchymal morphologies that strongly correlate with reporter fluorescence. Furthermore, we have shown that small molecule probes can effectively modulate EMT phenotypes producing reversion of the mesenchymal phenotype to an epithelial phenotype. Therefore, characterization of EMT phenotypes will enable us to mechanistically understand EMT plasticity driving malignant cancer and to develop therapeutic strategies to target EMT.

Topic II: "Deep learning for high-content screening and image-based profiling: potential and pitfalls"

Discussion Leader: Anne E. Carpenter, Ph.D. Institute Scientist, Senior Director, Imaging Platform, Merkin Institute Fellow, Broad Institute of Harvard and MIT

Proof of principle studies have shown that deep learning might improve many steps in microscopy-based experiments, from segmentation to phenotype classification to clustering samples by their image-based profile. In this discussion, participants will share their experiences in applying deep learning algorithms to their image-based research. We will discuss potential pitfalls and current limitations, as well as the tremendous potential.