TIME SERIES MULTIVARIATE CLUSTERING

In this project we used unsupervised machine learning to classify the behavior of engeneered T cells in a 3D in vitro assay.

Type of data: Imaging time series data from engeneered T cells in 3D co-culture with patient derived tumor organoids.

Background: In our assay we investigated how T cells where behaving during the course of tumor attack. In particular here we were interested in understanding heterogeneity of T cell behavior during this process.

Goal: We aimed to classify cells according to their behavioral characteristics (motility and interaction with other cells).

Approach: Using unsupervised machine learning (dynamic time warping algorithm) I separated cells into clusters based on their behavior.

Main result: We idenitfied a behavioral signature that was characteristic of cell that where responsable for actively killing tumor cells.