Machine Learning for Meaningful Lives

Welcome to the Machine Learning for Meaningful Lives Lab, also known as the M2Lab!

We develop advanced machine learning methods and apply them to biomedical research to uncover biological mechanisms, enhance disease detection, and support precision treatment selection. From the machine learning perspective, our work focuses on high-dimensional and small-sample challenges, optimization, structured variable selection, density estimation, active learning, multiple-instance learning, longitudinal analysis, and multimodal integration. We extend state-of-the-art frameworks such as contrastive learning, transformers, diffusion models, and Mamba neural networks, etc. From the bioinformatics and computational biology perspective, we investigate gene regulation (transcription and translation), heterochromatin dynamics, and automated feature extraction for time-series electrocardiology data and volumetric 3D cell imaging. In practice, these methods have been successfully applied to biomarker discovery, cancer detection, and protein synthesis estimation. These research topics can be broadly classified into two interconnected domains — medical digital twins and digital cells, powered by AI.

Medical digital twins

Digital twins are intelligent, adaptive computational replicas of individual patients that continuously evolve by integrating real-time, multimodal data streams—such as vital signs, laboratory results, medical imaging, genomics, and clinical narratives. These dynamic models emulate physiological processes, anticipate disease progression, and evaluate potential interventions before they are implemented in practice. By mirroring a patient’s unique and evolving health profile, digital twins empower clinicians to deliver personalized, data-driven care with greater precision and foresight. They serve as a foundation for next-generation precision medicine, enabling predictive analytics, continuous learning, and closed-loop feedback systems that connect real-world clinical data with virtual simulation environments for continual model refinement.

Digital cells

Digital cells are advanced computational representations of individual cells that integrate high-resolution single-cell and spatial multi-omics data to model cellular identity and behavior. By capturing gene expression, chromatin accessibility, protein profiles, and spatial context, digital cells reconstruct molecular mechanisms underlying development, disease progression, and therapeutic response. These models reveal dynamic lineage trajectories, cell–cell communication networks, and microenvironmental influences that shape cellular states. As digital counterparts to living cells, they enable in silico experimentation, virtual perturbation analysis, and predictive modeling — laying the foundation for precision diagnostics, targeted drug discovery, and the next generation of data-driven biomedical research.

For more details, please visit the Research Overview, Publications and Presentations, and Teaching and Advising pages, or explore our Principal Investigator’s introduction.

Google Sites

Report abuse