Stream 3: AI-Powered Therapeutic Design

Overview

Stream 3 closes the QuantOmics pipeline. Trainees here build the predictive AI models that translate the high-fidelity multi-omic datasets generated by Streams 1 and 2 into actionable therapeutic insights — identifying immunogenic targets, predicting treatment response, and guiding the rational design of personalized vaccines and precision therapeutics.

This stream operates at the intersection of machine learning, structural biology, and clinical translation. It is where the quantum-to-clinic vision of QuantOmics is most directly realized.

Research Focus Areas

Neoantigen Prediction & Vaccine Design

Personalized cancer vaccines require identifying tumor-specific neoantigens — mutated peptides that the immune system can be trained to recognize. Trainees work on:

• Benchmarking and adapting next-generation pathogenicity predictors to identify immunogenic neoantigens for personalized vaccine design
• Developing multimodal fusion models that integrate genomic variant data with structural protein features and HLA binding affinity predictions
• Building end-to-end vaccine design pipelines from quantum-enhanced sequencing data to candidate antigen ranking

Generative AI for Drug Discovery

Using generative models to explore vast chemical spaces and identify novel therapeutic candidates:

• Applying generative AI models (VAEs, diffusion models) for anomaly detection to identify drug-resistant cells in organoid cultures
• Designing latent-space representations of molecular structures conditioned on multi-omic signatures
• Generating and evaluating novel small molecule candidates for precision targeting

Multimodal Fusion for Clinical Diagnostics

Combining signals across modalities for robust clinical prediction:

• Creating deep learning frameworks for the multimodal fusion of electrical, optical, and genomic sensor data for cardiotoxicity screening
• Developing transformer-based models for multimodal representation learning across genomic, imaging, and clinical data
• Building uncertainty-aware prediction models for clinical deployment in high-stakes settings

Immune Cell Profiling & Treatment Efficacy

Quantifying the immune response is critical to evaluating therapeutic efficacy:

• Using advanced loss functions and segmentation models to precisely identify and quantify immune cell infiltration in microscopy images
• Providing a direct, AI-based measure of therapeutic efficacy in organoid and in vitro models
• Developing self-supervised AI models to automatically classify cellular phenotypes from high-content imaging data, enabling large-scale unbiased analysis

Trustworthy & Responsible Clinical AI

AI deployed in clinical settings must be fair, interpretable, and robust:

• Developing explainability frameworks (SHAP, integrated gradients, concept-based explanations) for genomic AI models
• Quantifying and mitigating performance disparities across patient demographic groups
• Building validation frameworks for quantum-AI systems navigating Health Canada regulatory pathways

Example Trainee Projects

• Personalized neoantigen vaccine pipeline — an end-to-end system from quantum-enhanced WGS of tumor organoids to a ranked list of candidate immunogenic peptides
• Generative molecular design for drug resistance — applying a diffusion model to identify novel inhibitor candidates for drug-resistant cancer cells detected in organoid assays
• Multimodal cardiotoxicity classifier — a fusion model combining quantum sensor electrical signals, microscopy images, and gene expression data to predict drug-induced cardiac toxicity
• Immune infiltration quantification tool — a deep learning segmentation system deployed on multiplexed fluorescence images of treated organoids to measure therapeutic response
• Bias audit framework for genomic AI — a systematic evaluation tool that measures and corrects performance disparities of genomic AI models across ancestrally diverse patient populations

The “Personalized Vaccine Design Challenge”

Stream 3 forms the intellectual core of QuantOmics’ signature annual Personalized Vaccine Design Challenge — a biennial team hackathon where interdisciplinary trainee teams tackle the complete sensor-to-vaccine pipeline, culminating in a presentation to a panel of academic, clinical, and industry judges. This challenge uniquely forces integration across all three streams.

Equity & Responsible AI in Therapeutic Design

QuantOmics explicitly trains Stream 3 trainees to address AI ethics in clinical applications. Key training elements include:

• Course 1.5 — Responsible Innovation & EDI in Precision Health: regulatory pathways, inclusive AI design, Indigenous data sovereignty
• Frameworks for ethical AI in oncology and emerging quantum technologies
• Proactive identification and correction of demographic biases at the model design stage, not as an afterthought

Stream 3 Co-Leads

• Dr. Naimul Khan (TMU) — Multimodal ML, AI in healthcare, biosensing (Program Director)
• Dr. Amber Simpson (Queen’s) — Biomedical computing, cancer informatics, AI for clinical data (Tier 2 CRC)

Contributing Faculty:

• Dr. Brenda Andrews (UofT) — Systems genetics, functional genomics (Tier 1 CRC)
• Dr. Jacques Corbeil (U Laval / MILA) — Genomic AI, multi-omics

Courses Supporting Stream 3

• Course 1.3 — AI in Genomics (primary course for this stream)
• Bootcamp 1.4 — Multimodal-Omics Data Integration
• Course 1.5 — Responsible Innovation & EDI in Precision Health