A Novel Computational Framework for Predicting Immunotherapy Response
An interactive overview of a postdoctoral research plan to develop a multi-modal deep learning tool for personalized cancer treatment.
The Clinical Challenge
Immune checkpoint inhibitors (ICIs) are a powerful cancer treatment, but their effectiveness is unpredictable. This research addresses the critical need for reliable biomarkers to guide therapy.
High Non-Response Rate
A significant portion of non-small cell lung cancer (NSCLC) patients do not respond to ICI therapy.
Elusive Biomarkers
Current predictive markers are insufficient, leading to a trial-and-error approach in treatment selection.
Increased Healthcare Costs
Ineffective treatments are costly and expose patients to unnecessary side effects.
An Integrative Deep Learning Approach
This project moves beyond single-data-type analysis by creating a framework that synergistically combines two powerful data sources to model the complex tumor microenvironment (TME). Hover over each component to learn more.
Specific Research Aims
The research is organized into two primary, interconnected aims that build upon each other to achieve the project's overall goal.
Aim 1: Model Development
Develop and optimize a graph neural network (GNN) architecture for the integration of whole-slide image (WSI) features and spatial transcriptomic data. This aim focuses on constructing a model that can learn from the complex cellular architecture of the TME. We will employ self-supervised learning techniques to extract hierarchical features from the WSIs and map them to spatially co-registered transcriptomic profiles to build a comprehensive, spatially-aware model of the TME.
Aim 2: Validation & Biomarker Discovery
Validate the predictive performance of the model on independent patient cohorts and identify novel, spatially-informed biomarkers of ICI response. The trained model will be validated using retrospective, multi-institutional cohorts with known clinical outcomes. Explainable AI (XAI) techniques will be applied to the validated model to deconstruct its decision-making process. This will enable the identification of key cellular morphologies and gene signatures that are most predictive of therapeutic response.
Validation and Potential Impact
The model's performance will be rigorously tested against clinical data, with the ultimate goal of creating a clinically translatable tool that benefits patients, clinicians, and the research community.
Illustrative ROC curve to evaluate model performance.
For Patients
Enables personalized treatment plans, increasing the likelihood of success and avoiding unnecessary side effects.
For Clinicians
Provides a robust decision-support tool to guide the selection of the most effective therapy for each patient.
For Research
Uncovers novel biological mechanisms and biomarkers of immunotherapy response and resistance.