Events Calendar

Uncovering Cellular Subtypes with AI-based Analysis of Heterogeneity

This is a past event.

Friday, December 1, 2023, 3 pm– 4 pm

This is a past event.

Biomedical Engineering Research Seminar

Kwonmoo Lee, PhD

Vascular Biology Program, Boston Children's Hospital
Department of Surgery, Harvard Medical School


In the pursuit of personalized medicine and understanding disease mechanisms, effective disease and cell subtyping are crucial. Single-cell data offers new opportunities, but its high-dimensionality pose challenges. While feature selection can streamline subtyping by reducing features, conventional methods often compromise heterogeneity, hindering subtype discovery. Our deep metric learning-based feature embedding, emphasizing significant differences in interquartile range (IQR), preserves heterogeneity while upholding discrimination among known cellular states. Leveraging this, we introduced PHet (Preserving Heterogeneity), a statistical method that identifies heterogeneity-preserving discriminative features for subtype clustering. PHet, validated on microarray and scRNA-seq datasets, outperformed previous methods. Furthermore, it revealed novel basal cell subtypes in airway epithelium and identified a subpopulation of breast cancer stem cells with distinctive motility and morphological characteristics.

Additionally, our innovation extends to HoloNet, a novel deep learning architecture for breast cancer cell phenotyping. Utilizing lens-free inline holography, HoloNet extracts large features from diffraction patterns, integrates them with convolutional layers, and surpasses existing methods in classifying breast cancer cells. This approach enhances interpretability of raw holograms, capturing markers like ER/PR and HER2. HoloNet's embedding further reveals shared breast cancer cell subtypes across diverse types, facilitating detailed analyses of rare and subtle phenotypes. This comprehensive strategy advances our ability to precisely diagnose breast cancer through an in-depth exploration of cell phenotype heterogeneity.


Dr. Kwonmoo Lee is an expert in machine learning and cell biology, with a particular focus on applying AI/machine learning and live cell imaging to decipher the complexities of cancer and regeneration. After earning his PhD in physics from MIT, Dr. Lee went on to complete postdoctoral training at Harvard Medical School. He then held an assistant professor position at Worcester Polytechnic Institute before joining the Vascular Biology Program at Boston Children's Hospital and the Department of Surgery at Harvard Medical School in 2020. Dr. Lee's research seeks to synergize the capabilities of AI/machine learning with live cell imaging to uncover phenotypic heterogeneity in cellular morphodynamics and motility. This exploration spans both healthy and pathophysiological conditions, with the ultimate goal of identifying innovative molecular targets for personalized medicine.

Stream Details

Additional Details



  • Quang Tran

1 person added

User Activity

No recent activity