We developed a containerized deep learning pipeline for automated segmentation and classification of tumor, immune, and stromal cells from whole-transcriptome spatial imaging data. By integrating per-cell transcriptomic, morphological, and spatial features, the framework generates harmonized, high-confidence single-cell annotations across sites. These outputs enable robust spatial and topological biomarker analyses to advance precision oncology.

MethylFM is a transformer-based foundation model to capture context-aware methylation patterns and to enable multiple downstream tasks. Our work addresses the challenge of leveraging WGBS data to uncover relationships between methylation, histone modifications, and cellular identity, with applications in disease biomarker discovery and therapeutic development.

As AI models (AlphaMissense, AlphaGenome etc) variant interpretation algorithms increasingly shape genomic research, their predictions remain probabilistic—powerful but unverified. This talk presents a practical framework for experimentally validating AI-predicted pathogenic variants using high-throughput splicing assays (HTSA), multimodal genomic data, and clinical correlations from 15,000+ patient registry. Attendees will learn how integrating AI-driven insights and real-world patient data transforms variant hypotheses into actionable, clinically credible conclusions.

Next-generation biomedical technologies for individualized diagnostics and longitudinal patient monitoring increasingly embed AI and machine learning directly into the devices to enhance accuracy, robustness, and clinical utility. By combining novel biomarker data from these sensors with hospital EHR systems, we can enable advanced, truly personalized patient monitoring platforms. This talk will highlight recent work from Dr. Hassan’s research laboratory in developing such AI-enabled biomedical sensing technologies.

This presentation introduces a biology-informed AI platform designed for scalable, multi-cancer early detection using blood-based molecular signals. By combining mechanistic priors with advanced machine learning, the platform improves sensitivity, reduces false positives, and enhances tumor-type localization across diverse populations. Attendees will learn how integrative feature modeling, robust validation workflows, and real-world deployment data support a clinically viable, population-scale approach to early detection.

AI in healthcare has accelerated dramatically, transforming radiology, genomics, and oncology, but psychiatry has remained largely untouched by this precision-medicine revolution. Existing suicide-risk algorithms focus on static, historical predictors and yield probabilistic scores that clinicians describe as “informative but not actionable.” At the same time, breakthroughs in liquid biopsies and mitochondrial DNA (cf-mtDNA) biomarkers are revealing that psychiatric crises may have measurable molecular signatures. Yet, no current platform meaningfully integrates biological and computational signals into a clinician-usable, real-time decision support system.

This talk introduces Proteus-AI, a new precision-behavioral-health framework designed to bridge that gap. This talk will synthesize current evidence from AI/ML in psychiatry, digital phenotyping, cf-mtDNA biomarker research, and implementation science. The Proteus-AI framework layers:
1) Explainable AI trained on longitudinal EHR trajectories; 2) Behavioral-health deterioration detection rather than static suicide prediction; and 3) Mitochondrial cell-free DNA and autoantibody biomarker signals to ground psychiatric risk in measurable biology.

The talk will examine how AI can be integrated into clinical workflow, how explainability builds clinician trust, and how ethical guardrails ensure responsible deployment among high-risk Veterans/civilians. By reframing AI not as a black box but as a transparent interventional tool. Proteus-AI aims to redefine precision psychiatry for suicide prevention, moving from prediction to prevention.

Audience members will leave with: 1) A clear view of global trends in psychiatric AI and precision biomarkers, 2) An implementation blueprint for integrating AI/ML into EHR workflows, and 3) A vision for how mental health can finally achieve what oncology has already demonstrated: precision medicine that saves lives.

As AI becomes embedded in radiology, triage, population health, and operational decision support, ensuring its clinical dependability is essential. This session introduces a predictive assurance framework that scores AI systems based on drift, anomaly likelihood, workflow sensitivity, and safety impact. The approach combines telemetry-driven monitoring with explainable auditing aligned to NIST AI RMF and emerging FDA expectations. Validated in enterprise healthcare environments, it improves early detection of reliability degradation and strengthens collaboration between data science, clinical, and compliance teams. Participants will learn actionable techniques for deploying trustworthy, regulated-grade AI.

Biomedical digital twins are redefining oncology and precision medicine by integrating multi-modal patient data, AI-driven modeling, and predictive analytics. This session explores how digital twins enhance disease prediction, treatment optimization, and clinical decision-making in oncology and biopharma. Learn how AI-enabled twin models are addressing data and validation challenges while paving the way for more adaptive, personalized, and patient-centered care pathways.

AI, genomics, and multimodal data science are reshaping rare-disease diagnosis and dramatically reducing the diagnostic odyssey. This closing joint session brings together leaders in precision medicine, bioinformatics, national rare-disease infrastructure, and real-world patient advocacy to highlight breakthrough models for rapid genomic interpretation, data integration, and clinical deployment. Attendees will gain a unified, cross-disciplinary view of what’s required to deliver faster, more accurate, and more equitable rare-disease diagnoses.