Kaizhong (Mike) Mu

Research Interests

AI + Statistics

Longitudinal Data

Neuroimaging & Network Data

Bayesian statistics

Publications & Manuscripts

Mu, K., Zhang, D., & Eloyan, A. A Reliable Change Estimation Method for Severely Cognitively Impaired Populations. Manuscript in preparation for submission to Alzheimer’s & Dementia. 1st author; draft completed July 2025, currently undergoing formatting.

Ryspayeva, D., Seyhan, A. A., Mu, K., Liu, M., Purcell, C., MacDonald, W. J., Halytskiy, V., Drevytska, T., Inomistova, M., Khranovska, N., Potorocha, O., Taran, L., Sumkina, O., Smolanka, I. Sr., & El-Deiry, W. S. Longitudinal miRNA Profiles in Breast Cancer Tissue and Plasma: Associations with Hormone Receptors, Response, and Survival. World Journal of Clinical Oncology. Manuscript submitted October 2025 (Manuscript No. 115287). 3rd & 4rd author.

Current & Past Research

An Bayesian-based RNN Method for MNAR/MAR + left-censored longitudinal

Advisor: Dr. Ani Eloyan
Duration: 09/25 – Present

This thesis extends prior SRB-based work by replacing its predefined linear trend and single-retest constraint with an RNN architecture to fully utilize multiple follow-ups while preserving Bayesian uncertainty quantification.
Constructed a unified likelihood that simultaneously handles MNAR/MAR missingness and left-censoring in Alzheimer’s longitudinal data through RNN hidden states that capture temporal dependence.
Implemented Bayesian inference, specifically variational inference, enabling the RNN to inherently model incomplete data while maintaining uncertainty quantification and computational efficiency.
Basic derivations completed, with manuscript preparation in progress; to be presented as a poster at Brown in April 2026

A Bayesian Cognitive Change Method for Censored Longitudinal Data

Advisor: Dr. Ani Eloyan
Duration: 01/25 – 07/25

Designed simulation study and an R Shiny App to show that SRB method underestimates cognitive change in Severe dementia whose follow-ups are left-censored at floor of NACC test battery in Longitudinal EOAD Study (LEADs)
Extended the SRB method by introducing Bayesian Inference that models censored follow-ups as random variables under a censored (Tobit-type) likelihood, realized through the MCMC (Hastings-Metropolis) implemented in R Stan.
Conducted simulation and real-data analyses demonstrating the proposed method reduce the underestimation while preserving the interpretability aspect of the SRB index and providing credible intervals for uncertainty quantification.
Resulted in a manuscript under formatting and presentations at three conferences.

Nonlinear Mixed Effect Model for Alzheimer’s Disease Biomarker

Advisor: Dr. Ani Eloyan
Duration: 09/24 – 12/24

Conducted literature reviews on vivo Alzheimer’s disease imaging biomarker (tau-PET, WMH in MRI) and Spline modeling techniques (B-spline, Natural-spline, penalized-spline)
Constructed a P-spline design matrix based on cubic B-spline basis function within mixed-effects framework to model the nonlinear increment of White Matter Hyperintensities (WMH) volume in longitudinal MRI data.

Translational Medical Research Collaboration

Collaborator: Dr. Dinara Ryspayeva, Dr. Attila Seyhan, Dr. Wafik El-Deiry. Legorreta Cancer Center
Duration: 03/25 – 09/25

Led and conducted a full pipeline of statistical analyses for breast cancer miRNA biomarker discovery —including missing data imputation, survival data collection guidance, survival analysis.
Identified MAR assumption via sensitivity analyses and applied conditional mean imputation for missing data
Built Cox regression model to evaluate the predictive associations of miR-34a, miR-137, miR-373, miR-124a, and miR-155 with overall survival, taking account for both biological interpretability (up/down regulation, interactions) and statistical rigor (PH assumption, multicollinearity, influential point).
Resulted in a manuscript currently under submission to “World Journal of Clinical Oncology”

Health Data Science Summer Fellowship Program

Advisor: Dr. Alyssa Bilinski
Duration: 06/25 – 08/25

Received training in data tidying in R, visualization in Tableau, data management in SQL, and causal inference.
Applied Large Language Model (LLMs) (OpenAI) for free-text classification of public response to private equity, aiming to evaluate and improve prompt effectiveness in domain-specific text understanding.
Built an automated evaluation pipeline and interactive Shiny App to benchmark AI annotations against human ground truth.
Resulted in oral presentation to program faculty and fellows at the final symposium.

Comparative Analysis of Machine Learning Models for Unemployment Prediction

Advisor: Dr. Colin Cameron
Duration: 12/22 – 03/23

Collected and managed unemployment related data through APIs, web scraping (BeautifulSoup) in Python
Implemented and Compared multiple regression models, including OLS, LASSO, and a novel dimension-reduced local weighted regression applying PCA, for unemployment rate prediction.
Evaluated and visualized model results in Stata and RStudio, using key metrics like Mean Squared Error (MSE), Mean Absolute Deviation (MAD), and R-squared, accuracy, and efficiency.

Graph-Based Change-Point Detection Method for High-Dimensional Network Data

Advisor: Dr. Hao Chen, Assoc. Prof. of Statistics, UCD
Duration: 09/22 – 01/23

Conducted literature reviews on graph theory and graph models (Stochastic Block Models, Erdős–Rényi graphs and Configuration Model)
Refined gSeg R-package, implementing a combination of Image Analysis and SCAN statistics test in detecting change-point in dynamic social networks, while optimizing the algorithms for computational efficiency and scalability
Designed and conducted simulation studies in R to evaluate the predictive performance of the proposed method against conventional approaches, across different levels of graph sparsity, edge density, and network connectivity

Presentations and Conferences

Date	Conference	Talk & Award
Jun 9-11, 2025	Southern Regional Council On Statistics (SRCOS)	Poster presentation & Boyd Harshburger Travel Award Winner
Aug 14-16, 2025	Statistics in Pharmaceuticals (SIP)	Poster presentation
Oct 10, 2025	ASA NJ Statistics Workshop 2025	Poster presentation

--- listing: - id: phdrs contents: "/research/" type: grid grid-columns: 2 grid-item-align: center image-lazy-loading: true fields: [title, image, author,date] sort: "date desc" --- <script async src="https://pagead2.googlesyndication.com/pagead/js/adsbygoogle.js?client=ca-pub-6878992848042528" crossorigin="anonymous"></script> ## Research Interests ::: {#phdrs} ::: ## Publications & Manuscripts **Mu, K.**, Zhang, D., & Eloyan, A. A Reliable Change Estimation Method for Severely Cognitively Impaired Populations. Manuscript in preparation for submission to Alzheimer's & Dementia. 1st author; draft completed July 2025, currently undergoing formatting. Ryspayeva, D., Seyhan, A. A., **Mu, K.**, Liu, M., Purcell, C., MacDonald, W. J., Halytskiy, V., Drevytska, T., Inomistova, M., Khranovska, N., Potorocha, O., Taran, L., Sumkina, O., Smolanka, I. Sr., & El-Deiry, W. S. Longitudinal miRNA Profiles in Breast Cancer Tissue and Plasma: Associations with Hormone Receptors, Response, and Survival. World Journal of Clinical Oncology. Manuscript submitted October 2025 (Manuscript No. 115287). 3rd & 4rd author. ## Current & Past Research ### An Bayesian-based RNN Method for MNAR/MAR + left-censored longitudinal **Advisor:** Dr. Ani Eloyan **Duration:** 09/25 – Present - This thesis extends prior SRB-based work by replacing its predefined linear trend and single-retest constraint with an RNN architecture to fully utilize multiple follow-ups while preserving Bayesian uncertainty quantification. - Constructed a unified likelihood that simultaneously handles MNAR/MAR missingness and left-censoring in Alzheimer's longitudinal data through RNN hidden states that capture temporal dependence. - Implemented Bayesian inference, specifically variational inference, enabling the RNN to inherently model incomplete data while maintaining uncertainty quantification and computational efficiency. - Basic derivations completed, with manuscript preparation in progress; to be presented as a poster at Brown in April 2026 ### A Bayesian Cognitive Change Method for Censored Longitudinal Data **Advisor:** Dr. Ani Eloyan **Duration:** 01/25 – 07/25 - Designed simulation study and an R Shiny App to show that SRB method underestimates cognitive change in Severe dementia whose follow-ups are left-censored at floor of NACC test battery in Longitudinal EOAD Study (LEADs) - Extended the SRB method by introducing Bayesian Inference that models censored follow-ups as random variables under a censored (Tobit-type) likelihood, realized through the MCMC (Hastings-Metropolis) implemented in R Stan. - Conducted simulation and real-data analyses demonstrating the proposed method reduce the underestimation while preserving the interpretability aspect of the SRB index and providing credible intervals for uncertainty quantification. - Resulted in a manuscript under formatting and presentations at three conferences. ### Nonlinear Mixed Effect Model for Alzheimer's Disease Biomarker **Advisor:** Dr. Ani Eloyan **Duration:** 09/24 – 12/24 - Conducted literature reviews on vivo Alzheimer's disease imaging biomarker (tau-PET, WMH in MRI) and Spline modeling techniques (B-spline, Natural-spline, penalized-spline) - Constructed a P-spline design matrix based on cubic B-spline basis function within mixed-effects framework to model the nonlinear increment of White Matter Hyperintensities (WMH) volume in longitudinal MRI data. ### Translational Medical Research Collaboration **Collaborator:** Dr. Dinara Ryspayeva, Dr. Attila Seyhan, Dr. Wafik El-Deiry. Legorreta Cancer Center **Duration:** 03/25 – 09/25 - Led and conducted a full pipeline of statistical analyses for breast cancer miRNA biomarker discovery —including missing data imputation, survival data collection guidance, survival analysis. - Identified MAR assumption via sensitivity analyses and applied conditional mean imputation for missing data - Built Cox regression model to evaluate the predictive associations of miR-34a, miR-137, miR-373, miR-124a, and miR-155 with overall survival, taking account for both biological interpretability (up/down regulation, interactions) and statistical rigor (PH assumption, multicollinearity, influential point). - Resulted in a manuscript currently under submission to "World Journal of Clinical Oncology" ### Health Data Science Summer Fellowship Program **Advisor:** Dr. Alyssa Bilinski **Duration:** 06/25 – 08/25 - Received training in data tidying in R, visualization in Tableau, data management in SQL, and causal inference. - Applied Large Language Model (LLMs) (OpenAI) for free-text classification of public response to private equity, aiming to evaluate and improve prompt effectiveness in domain-specific text understanding. - Built an automated evaluation pipeline and interactive Shiny App to benchmark AI annotations against human ground truth. - Resulted in oral presentation to program faculty and fellows at the final symposium. ### Comparative Analysis of Machine Learning Models for Unemployment Prediction **Advisor:** Dr. Colin Cameron **Duration:** 12/22 – 03/23 - Collected and managed unemployment related data through APIs, web scraping (BeautifulSoup) in Python - Implemented and Compared multiple regression models, including OLS, LASSO, and a novel dimension-reduced local weighted regression applying PCA, for unemployment rate prediction. - Evaluated and visualized model results in Stata and RStudio, using key metrics like Mean Squared Error (MSE), Mean Absolute Deviation (MAD), and R-squared, accuracy, and efficiency. ### Graph-Based Change-Point Detection Method for High-Dimensional Network Data **Advisor:** Dr. Hao Chen, Assoc. Prof. of Statistics, UCD **Duration:** 09/22 – 01/23 - Conducted literature reviews on graph theory and graph models (Stochastic Block Models, Erdős–Rényi graphs and Configuration Model) - Refined gSeg R-package, implementing a combination of Image Analysis and SCAN statistics test in detecting change-point in dynamic social networks, while optimizing the algorithms for computational efficiency and scalability - Designed and conducted simulation studies in R to evaluate the predictive performance of the proposed method against conventional approaches, across different levels of graph sparsity, edge density, and network connectivity ## Presentations and Conferences | Date | Conference | Talk & Award | |------|------------|--------------| | Jun 9-11, 2025 | Southern Regional Council On Statistics (SRCOS) | Poster presentation & Boyd Harshburger Travel Award Winner | | Aug 14-16, 2025 | Statistics in Pharmaceuticals (SIP) | Poster presentation | | Oct 10, 2025 | ASA NJ Statistics Workshop 2025 | Poster presentation |