Min Woo Sun

I'm a PhD candidate at Stanford working on machine learning for biology and medicine. I'm fortunate to be jointly advised by Serena Yeung-Levy and Robert Tibshirani and to be supported by the National Library of Medicine T15 grant and the Stanford Data Science Scholars program.

Most recently, I interned as a Machine Learning Engineer at Hugging Face. In the past, I worked on NGS lab workflows at Invitae and early cancer detection at Guardant Health.

Email  /  Scholar  /  Github  /  LinkedIn  /  🤗

Research

I'm interested in vision language models, computer vision, and statistical machine learning.

We introduce BIOMEDICA, an open-source framework that transforms the PubMed Central Open Access subset into a comprehensive dataset of over 24 million image-text pairs with expert-guided annotations, enabling state-of-the-art performance in biomedical vision-language models across diverse tasks and domains.

Functions to summarize DNA methylation data using regional principal components. Regional principal components are computed using principal components analysis within genomic regions to summarize the variability in methylation levels across CpGs.

We developed a computer vision model to analyze laparoscopic surgery videos, identifying fine-grained surgical actions linked to operative blood loss and surgeon experience.

Cooperative Component Analysis (CoCA) is an unsupervised multi-view analysis method that balances within-view variance preservation and cross-view correlation through a trade-off between approximation error and agreement penalty. It generalizes PCA and CCA, with a sparse variant enabling feature selection, and demonstrates effectiveness in integrating multiomics data for disease progression prediction.

Molecular data from mass spectrometry were used to build classifiers, achieving high diagnostic accuracy when compared to pathology results, highlighting its potential for real-time surgical guidance.

Cross-validation (CV) can underestimate test error variance due to correlated error estimates from using the same samples for training and testing. Nested CV mitigates this issue by providing more accurate coverage through improved error variance estimation, which this work extends to the Cox proportional hazards model.

The COVID-19 pandemic has caused 14.9 million excess deaths globally, with prior studies linking cross-country differences in COVID-19 deaths to demographic and health factors. This analysis extends the scope by incorporating government policies, showing the critical role of public health efforts in reducing excess deaths.

A therapeutic smartphone game, GuessWhat, was used to collect and label a large dataset of child emotion expressions, enabling the training of a CNN classifier with significantly improved accuracy (up to 79.1% balanced accuracy) compared to prior models. This approach shows the potential of gamified data collection for enhancing emotion recognition in pediatric digital health care.

We introduce game theoretic centrality, which integrates biological network knowledge with Shapley value from coalitional game theory to prioritize disease-associated genes. Applied to autism spectrum disorder (ASD), the approach identifies biologically relevant genes, demonstrating potential regulatory interactions and offering insights into the genetic basis of complex disorders.