Classifying Metamorphic versus Single-Fold Proteins with Statistical Learning and AlphaFold2
Annals of Applied Statistics, accepted, 2026
Statistical learning and AlphaFold2 features for identifying proteins with alternative folds.
Yongkai Chen 陈永凯
Postdoctoral Researcher
Department of Statistics
Harvard University
I am a postdoctoral researcher in the Department of Statistics at Harvard University, working with Samuel Kou. I received my Ph.D. in Statistics from the University of Georgia in 2024, advised by Wenxuan Zhong and Ping Ma, and my B.S. in Mathematical Statistics from Nankai University in 2019.
Research overview
My research develops statistical foundations for AI-driven scientific discovery, with applications in computational biology and protein science. My interests include Bayesian nonparametric modeling and inference, uncertainty quantification, and scalable and quantum statistical computing.
Topics:
Uncovering Distinct Protein Conformations Using Coevolutionary Information and AlphaFold
bioRxiv, 2025
SMICE uses coevolutionary information to guide MSA sampling and explore distinct protein conformations.
Wahkon: A Statistically Principled Deep RKHS Superposition Network
STAI-X, 2026 · Best Paper Award
Smooth link functions, RKHS regularization, and hierarchical Gaussian-process priors for deep learning.
Bayesian Knowledge Distillation: A Bayesian Perspective of Distillation with Uncertainty Quantification
ICML, 2024 · * Equal contribution
Teacher-informed priors connect knowledge distillation with Bayesian inference and uncertainty quantification.
Bisection Grover’s Search Algorithm and Its Application in Analyzing CITE-seq Data
Journal of the American Statistical Association, 2024 · * Equal contribution
Quantum search for statistical computation, with an application to multimodal single-cell data.
Quantum Statistical Bootstrap
arXiv preprint, 2026
Quantum algorithms for bootstrap computation, with an analysis of their statistical error properties.
Fluid Correlation: A Novel Nonparametric Metric to Assess the Dynamic Association
Journal of Computational and Graphical Statistics, 2025
A nonparametric metric for assessing dynamic association with Bayesian uncertainty quantification.
Orthogonal Multimodality Integration and Clustering in Single-Cell Data
BMC Bioinformatics, 2024 · * Equal contribution
OMIC integrates RNA and protein measurements to identify cell populations and interpret the contribution of each modality.
SpaDiff denoises sequence-based spatial transcriptomics via diffusion process
Cell Reports Methods, 2026
SpaDiff reverses RNA spot-swapping to recover spatial gene-expression patterns while preserving total molecular counts.
Protein science
Proteins are dynamic molecules whose changing shapes are closely connected to their biological roles. My research brings statistics and AI together to explore the diversity of protein structures and how proteins move between different conformations. I am interested in turning computational predictions into reliable insights about protein behavior.
Single-cell and spatial omics
Individual cells differ in their molecular activity, biological roles, and responses to their surroundings. My research develops statistical approaches to bring together complementary measurements of cells and understand how they are organized within tissues. The goal is to reveal meaningful cellular variation while accounting for the noise and complexity of modern biological data.
Multimodality integration: [paper]
SpaDiff: [paper]
BGS for CITE-seq: [paper] [code]
Microbiome
Microbial communities are dynamic ecosystems in which relationships between organisms can change over time and across environments. My research studies these changing associations to better understand the collective behavior of microbial communities. Through Fluid Correlation, I develop statistical tools that describe how these relationships evolve and quantify uncertainty in the resulting conclusions.
Fluid Correlation: [paper]
Biomedical imaging
Medical images offer a window into the structure and function of the human body, but translating them into useful information requires careful analysis. In collaborative work, I use statistics and machine learning to extract information from cardiac ultrasound and reconstruct three-dimensional heart morphology. These projects connect advances in data analysis with questions about cardiac structure and function.
S2MNet: [paper]
Neuroscience
Brain function emerges from the coordinated activity of many interacting neurons. My collaborative research uses statistical analysis to study how neural activity is organized and how connections between neurons change during development. By examining patterns across neural populations, we seek to understand how changes in biological conditions are reflected in the behavior of neural circuits.
Research implementations and analysis workflows associated with my papers.
- SMICE
- Coevolution-guided MSA sampling for exploring protein conformations with AlphaFold.
- Metamorphic-Classify
- Statistical learning workflows for classifying metamorphic and single-fold proteins.
- Wahkon
- A statistically principled deep RKHS superposition network.
- BGS for CITE-seq
- Bisection Grover’s search and its application to CITE-seq data analysis.
- Quantum Bootstrap
- Quantum algorithms and experiments for bootstrap computation.
- Neural Activity Analysis
- Analysis code for collaborative research on neural connectivity.
News
- Classifying Metamorphic versus Single-Fold Proteins accepted by the Annals of Applied Statistics.
- Wahkon received a Best Paper Award at STAI-X.
- Bisection Grover’s Search published in JASA. UGA news coverage.
- I completed my Ph.D. in Statistics at the University of Georgia and received the Best Senior Ph.D. Student award.
- I received a Summer Research Grant from the University of Georgia.
- I received the Best Beginning Ph.D. Student award at the University of Georgia.