Team 8: Differentiation of metastatic potential in melanoma cells

Early identification and characterization of melanoma is critical for successful treatment. Developing an assay using unperturbed patient derived cells to define a melanoma population’s metastatic characteristics would provide physicians with a powerful tool for planning therapeutic approaches. As such, it has been long speculated that metastatic cells adopt distinct morphodynamic states that are predictive of the metastatic potential of their tumor of origin.

We aim to develop an assay that can differentiate between patient derived xenograft melanoma cells with high and low metastatic properties.  Your mission, should you choose to accept it, is to design/build the computational analysis component of our assay that will analyze live cell movies using machine learning and computer vision techniques to predict metastatic status.

An experimental and imaging pipeline to harvest, culture, and live image (using label-free phase contrast microscopy in conditions that mimic the tumor microenvironment) primary human melanoma originating from 8 patients has been established. For this hackathon, approximately 12,000 individual live cell movies are available to analyze, with variable number of frames (up to 240 frames), and including non-tumor cells for reference (cell lines).  Each cell movie also contains important metadata, including the metastatic potential (high, low, n/a).

Distinguishing metastatic potential is the ultimate goal, yet a number of secondary quantitative analyses may be of interest, including clustering cells (or even distinct parts within a single movie of one cell) according to observed morphodynamic behavior/states and discerning cell-lines from primary tumors.

Team Lead: Andrew Jamieson, Bioinformatics,