Pathomics 9
Pathomics: Digital Pathology → Molecular Inference (GenAI + Foundation Models + Agentic RAG)
Predict gene and protein expression directly from H&E pathology slides using foundation and generative models.
Mentor Details:
Prof. Ilan Tsarfaty
Mentor Details:
Requirments:
Integrate pathology pipelines with generative molecular prediction models and validate against ground truth datasets.
Problem Statement
Molecular profiling is expensive and slow compared to histopathology imaging.
Project Objectives
Integrate pathology pipelines with generative molecular prediction models and validate against ground truth datasets.
Technical Scope
Integrate state-of-the-art pathology pipelines to predict gene/protein expression patterns directly from H&E tumor slides. You will run existing lab pipelines (tiling/normalization/feature extraction) and add modern generative molecular-prediction models, then validate outputs against available ground truth datasets.
Required Knowledge and Prerequisites
Core Requirements:
Python, ML fundamentals, image processing.
Recommended Background:
Computational pathology, microservices.
Project Difficulty and Expected Level
Overall Difficulty: High
This project is well-suited for:
Teams of 2–4 students
Expected Outcomes
End-to-end exposure to contemporary computational pathology: data curation, model integration, evaluation of metrics, and visualization.
“Platform engineering” skills: modularization into microservices, containerization, and Kubernetes-style deployment patterns used in real clinical AI stacks.
A strong applied-ML portfolio: multimodal evaluation (image → molecular), reproducibility, and rigorous reporting.