A New Representation for Single Cell Rna-sequencing Data and Covid-19 Data Analysis
Year: 2024
Type: dissertation
Abstract: An important pre-processing step in single cell RNA-sequencing (scRNA-seq) is the correction of various types of amplification noise and bias, such as the well-known library size effects. In traditional contexts, such as in bulk RNA-seq data, the amplification effect correction can be easily handled by appropriate multiplication. However, when there are a large number of zeros, such as in scRNA-seq data, simple multiplication is often ineffective, motivating new statistical approaches. We propose shifting the focus from modeling counts to modeling probabilities. A natural physical model is independent Poisson distributions with a different parameter for each entry of the data matrix. We investigate the validity of this model using a homogeneous single cell line experiment. Then based on such Poissoneity, we propose a novel data representation based on departures from estimated Poisson distributions for cell clustering as downstream analysis. Investigations using real data and crafted experiments show that compared with the popular Seurat approach, we are able to improve clustering performance in some situations. The second project studies some properties of coronavirus disease (COVID-19) infected data. Widely discussed superspreader events suggest that the tails of the daily infection numbers could be heavy, which is the main focus of our investigation. For each state in the United States, the mean-to-variance relationship shows that the mean of the daily-infected number is much smaller than the local variance. This indicates that the data are far from a standard Poisson process, and models based on the Poisson distribution may need substantial modification. After scaling and deseasonalizing, the time series show an impressive stationary property, which is appropriate for further time series analysis. Study of the marginal distributions of these time series fails to find tails as heavy as the Pareto distribution. The third project LCCC1424 is an observational, prospective cohort study of Kaposi Sarcoma (KS) patients treated at Kamuzu Central Hospital in Lilongwe, Malawi between February 2017 and June 2019. The central hypothesis is that treatment approaches for KS patients can be individualized if clinically relevant KS subsets can be defined. We define the baseline clinical characteristics of KS patients, and assess the response to the national standard of care by combining cART with bleomycin and vincristine chemotherapy at 48 weeks. The results from RNA-seq and differential gene expression analysis indicate that at least two potential subgroups of KS lesions exist within the HIV-positive KS population. Stratifying subsets of KS may help the development of more effective treatment strategies.
DOI: 10.17615/zsxt-7f60
Author: Pan, Yue
Institution: University of North Carolina at Chapel Hill
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W6962594741
Cites:
Cited by:
Related to: