Abstract 4306: Label-free melanoma phenotype classification using AI-based morphological profiling

Abstract Melanoma plasticity and heterogeneity contribute to therapeutic resistance and mortality, and the ability of melanoma cells to switch from melanocytic to mesenchymal phenotypes results in increased invasion and metastasis. Investigating the role of these phenotypic states has been challenging due to limited marker availability for each cell state. High-parameter molecular methods such as RNAseq have produced more descriptive gene signatures of phenotypic states, but these methods are cell-destructive, labor-intensive, expensive, and can take days to weeks to obtain a readout. Different morphologies have been observed for melanocytic and mesenchymal phenotypic states in culture, thus we hypothesized that morphology could provide an orthogonal method to define and study phenotypes. We utilized the Deepcell platform to develop a classifier to predict the phenotype of melanoma cells as melanocytic or mesenchymal based on morphology alone. This method is free of labels and utilizes viable cells, overcoming the technical and practical limitations of traditional, marker-based methods. We used 20 patient-derived cell lines with known melanocytic or mesenchymal transcription scores to develop the ‘Melanoma Phenotype Classifier’ to phenotype melanoma cells based on morphology alone. A classifier accuracy of 88% was achieved, and morphological analysis of the images revealed distinct morphotypes for each phenotype, highlighting distinct morphological differences. To further link phenotypic state with multi-dimensional morphological profiles, we performed genetic and chemical perturbations known to shift the phenotypic state. The Melanoma Phenotype Classifier successfully predicted shifts in phenotypes driven by these perturbations. These results further demonstrate how phenotype is linked to distinct morphological changes that are detectable by artificial intelligence. Lastly, we applied the Melanoma Phenotype Classifier to dissociated melanoma biopsies, which revealed phenotypic heterogeneity that was confirmed by single cell RNASeq. This work establishes a link between cellular morphology and melanoma phenotype and lays the groundwork for the use of morphology as a label-free method of phenotyping viable melanoma cells. Citation Format: Evelyn Lattmann, Andreja Jovic, Julie Kim, Aizhan Tastanova, Tiffine Pham, Christian Corona, Kiran Saini, Zhouyang Lian, Senzeyu Zhang, Ryan Carelli, Kevin B. Jacobs, Manisha Ray, Vivian Lu, Stephane C. Boutet, Mahyar Salek, Maddison Masaeli, Mitchell P. Levesque. Label-free melanoma phenotype classification using AI-based morphological profiling abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 4306.