Cellomics 8
Cellomic: Cancer Cell Motility & Metastasis (Foundation Models + Multivariate Time Series)
This project focuses on modeling cancer cell motility and metastatic transitions using large-scale time-lapse microscopy data and multivariate time-series representations.
Mentor Details:
Prof. Ilan Tsarfaty
Mentor Details:
Requirments:
Design AI pipelines for segmentation, tracking, trajectory modeling, and interpretable phenotyping to uncover metastasis-relevant motility programs.
Problem Statement
Cancer metastasis emerges from complex single-cell behaviors that are difficult to quantify due to noisy, high-dimensional imaging data.
Project Objectives
Design AI pipelines for segmentation, tracking, trajectory modeling, and interpretable phenotyping to uncover metastasis-relevant motility programs.
Technical Scope
Work with large-scale time-lapse microscopy data to model how cancer cells move, interact, and transition into metastatic phenotypes. You will convert single cell tracks into multichannel multivariate time series (MTS) and build AI pipelines for segmentation, tracking, behavior discovery, and downstream biological interpretation.
Required Knowledge and Prerequisites
Core Requirements:
Python, basic machine learning, computer vision fundamentals, data handling.
Recommended Background:
Deep learning, bioimaging, time-series analysis.
Project Difficulty and Expected Level
Overall Difficulty: varies
This project is well-suited for:
Teams of 2–4 students
Expected Outcomes
Practical experience with “real” bioimaging big data (noise, drift, missing frames, batch effects) and how to engineer robust pipelines.
Modern AI skills: foundation-model workflows for segmentation/tracking, unsupervised representation learning, clustering of trajectories, and interpretable phenotyping.
Production-oriented know-how: organizing data in MongoDB, exposing analysis through APIs/GUI, and building reproducible experiments.