MLx Health & Bio 2026

This module explores how to build AI systems that are more reliable for health and biomedical use. Learners will examine why robust performance matters beyond benchmark accuracy, including prediction instability across training samples and changes between model versions. The module introduces regularization as a way to reduce variance and improve consistency in real-world clinical settings. It also connects robustness to the broader requirements for regulated medical devices, including intended purpose, clinical evidence, quality management, and approval pathways.

1. 1. Building More Robust AI

This module introduces geometric deep learning for biomedical modelling and simulation, building on the challenges of AI in medical imaging. It explains why medical data are harder than natural images and how representing organs as meshes and graphs can better capture anatomy, topology, and physical behavior. Learners will compare convolutional, graph-based, and transformer methods, and see how physics-informed losses improve realism, robustness, and clinical relevance in tasks such as segmentation, registration, and simulation.

1. 1. Pre-reading: AI Applications in Medical Imaging
2. 2. Geometric Deep Learning For Biomedical Modelling & Simulation

This module introduces the Human Cell Atlas as a global effort to map human cells using single-cell and spatial genomics. Learners will explore how data science and AI enable cell type discovery, atlas assembly, multimodal integration, and reference building across organs. The module also examines workflows for clustering and annotation, batch correction, and linking genetic variants to cell-specific effects, with applications in tissue engineering, disease interpretation, and virtual cell and tissue modeling across lung, heart, brain, and thymus research.

1. 1. Human Cell Atlas, Data Science & AI

This module explores whether AI alone can transform disease treatment, using cardiovascular disease to show why computational medicine often needs more than data-driven models. Learners will examine how AI, physics-based modeling, and experimental evidence work together across scales to support digital twins, in silico trials, and safer therapy development. The module also highlights validation, regulatory credibility, and the importance of context of use, while explaining why simulation-first approaches are harder to adopt in medicine than in engineering.

1. 1. Can AI Alone Revolutionize the Fight Against Disease?

This module introduces AI for healthcare data through epidemiology and real-world clinical research. Using cardiovascular disease as a foundation, learners explore how electronic health records can estimate disease burden, analyze risk factors, support survival modeling, and evaluate interventions. The module also covers deep learning for prediction, model calibration, transfer learning across health systems, explainability, and causal inference. Case studies on statins, anticoagulants, Parkinson’s disease, heart failure, and target trial emulation highlight both the potential and limitations of AI in healthcare.

1. 1. AI for Healthcare Data

This module explores how artificial intelligence is being used at the bedside to improve cardiovascular care. Learners will examine the burden of heart disease and disparities in outcomes, with attention to women’s and maternal heart health. The module reviews AI-enabled ECGs, portable devices, echocardiography, and large language models, focusing on validation, clinical evidence, and implementation challenges. Through examples from Mayo Clinic and global collaborations, participants will see how these tools can support screening, diagnosis, risk stratification, and more efficient clinical workflows across diverse care settings.

1. 1. Artificial Intelligence at the Bedside

This module traces the evolution of machine intelligence from early pioneers like Turing and Shannon to deep learning and foundation models, then shows how these ideas are applied in medicine. Learners will explore clinical uses in brain and ovarian cancer, including medical image registration, brain shift during surgery, and multimodal data integration. The module also examines the practical barriers that shape real-world healthcare AI, helping learners connect core AI concepts with translational research and deployment challenges in clinical settings.

1. 1. Learning Machines; And What They Can Do For Medicine

This module introduces geometric deep learning as a symmetry-driven approach to machine learning for health and bio applications. Learners explore how invariance and equivariance shape model design, why inductive biases improve learning with limited data, and how convolution and message passing emerge from symmetry principles. The module also connects graph neural networks to graph isomorphism and expressive power, building intuition for modeling images, molecules, and biological networks.

1. 1. Geometric Deep Learning

This module introduces AlphaGenome, a unified DNA sequence model for predicting regulatory variant effects across human and mouse genomes. Learners explore how sequence inputs are used to infer regulatory activity, expression, splicing, accessibility, and chromatin contacts, along with the biological data that supervise training. The module also examines the hybrid convolution-transformer architecture, distillation strategy, benchmark evaluation, and practical workflows for scoring variants. Real-world applications and key limitations, including indels, distal regulation, and cell-type specificity, are highlighted.

1. 1. AlphaGenome: Advancing Regulatory Varient Effect Prediction with a Unified DNA Sequence Model

This module explores whether AI can transform drug discovery by grounding machine learning methods in the realities of biological targets, lead finding, and clinical development. Learners will examine molecular representations, virtual screening, docking, and protein structure prediction, including AlphaFold, while also assessing benchmark design, dataset bias, and model evaluation. The module highlights where AI is already delivering value, where limitations remain, and how to judge the promise of generative and predictive approaches in research and industry.

1. 1. Can AI Transform Drug Discovery?

This module introduces machine learning for routine health settings, with a focus on electronic health records, prognosis tasks, and the realities of large-scale hospital and insurance data. Learners explore tabular modeling approaches, missing-value handling, table foundation models, and multimodal stacking, while also addressing privacy, sparsity, and deployment constraints. The module concludes with methods for managing sampling bias, censoring, survival analysis, and causal inference so predictions remain useful in real-world healthcare.

1. 1. AI for Routine Health

Foundation Models for Wearables introduces how modern AI models learn from biosignals such as PPG, ECG, accelerometer, CGM, and sleep data. Learners will explore why wearable data is challenging to model, compare historical and self-supervised approaches, and examine key design choices in representations, objectives, architectures, and evaluation. The module also covers real-world health applications, along with privacy, fairness, and deployment considerations, preparing learners to assess how foundation models can move from prediction toward clinically meaningful action.

1. 1. Foundation Models for Wearables