Thanh M. Brown
BioInformatics, HealthTech, ML, NLP
Welcome to my pages
Bioinformatics
ISCVAM - Interactive Single-cell Visual Analytics tools for Multiomics
- We built ISCVAM as an interactive visual analytics tool for single-cell multiome. Our single-cell multiomics visualization app serves as an investigation tool, which allows users to:
(1) visualize data under multi-modalities (RNA, ATAC) --> leverage the transcriptomics and epigenetics of the cells
(2) paint cells with multiple single-cell clustering resolutions --> discover rare cell population
(3) view 3 datasets simultaneously --> validate findings across datasets
- ISCVAM can be accessed here: https://chenlab.chpc.utah.edu/iscvam/
- Poster was accepted to present at AACR 2023: https://aacrjournals.org/cancerres/article/83/7_Supplement/2086/722119/Abstract-2086-ISCVAM-an-Interactive-Single-Cell
(1) visualize data under multi-modalities (RNA, ATAC) --> leverage the transcriptomics and epigenetics of the cells
(2) paint cells with multiple single-cell clustering resolutions --> discover rare cell population
(3) view 3 datasets simultaneously --> validate findings across datasets
- ISCVAM can be accessed here: https://chenlab.chpc.utah.edu/iscvam/
- Poster was accepted to present at AACR 2023: https://aacrjournals.org/cancerres/article/83/7_Supplement/2086/722119/Abstract-2086-ISCVAM-an-Interactive-Single-Cell
Data Science
Osteoporosis prediction using machine learning
In this project, I performed EAD on a case-control dataset for osteoporosis, controlling for demographic variables such as age, gender, and ethnicity. Then I built predictive models using LR, RF, SVM, and XGB. The Gradient Boosting Classifier and Logistic Regression achieved the highest ROC scores. Even though this is a balance dataset, the variables tend to be bias towards the negative class (no-osteoporosis), meaning the True Negative and False Negative scores are always higher than True Positive and False Positive scores. This trend is consistent in the confusion matrix of the 6 ML models.
COVID-19 Vaccine adverse symptoms in VAERS with Association Rule Mining
To address vaccine hesitancy issues, I studied the adverse symptoms in COVID-19 vaccines using the VAERS data. VAERS is the public surveillance system co-manage by the CDC and FDA to detect rare vaccine adverse events. By applying association rule mining, I discovered the top adverse symptoms in COVID-19 vaccines, and compared the differences in adverse symptoms between Moderna and Pfizer vaccines.
Visualization app in RShiny using MAUDE dataset
MAUDE is the medical device passive surveillance dataset from the FDA. I built a visualization app to project the temporal trends in the harm levels using the 2016 MAUDE data.