Friday, Dec 3 -- Holt 171, 4:00pm
Jane Guo (Chico State University)
Title: Cluster Analysis Of US States Based On COVID-19 Incidence, Fatality, and Vaccination
Abstract: The recent COVID-19 pandemic has caused the loss of many lives worldwide. In the US, as of April 10, 2021, there have been more than 30 million accumulative confirmed cases and more than 550,000 reported COVID-19 related fatalities. Existing published studies presented the clustering results based on incidence and mortality information at different levels, including country and county levels. With the availability and accessibility of the COVID- 19 vaccines, it is essential to characterize the clustering features at the state- level using the information including the number of COVID-19 cases per 100 thousand in the last seven days, testing positivity rate in the last seven days, deaths per 100k in the last seven days as well as the vaccination rates. In the present study, we use principal component analysis to extract features from highly correlated data such as state-level incidence and mortality, as well as vaccination. We further performed hierarchical cluster analysis based on extracted features from state-level COVID-19 data. The presented cluster results could contribute to the pandemic management, including the policy- making process and health education program.
Friday, Oct 15 -- Holt 171, 4:00pm
Ricardo Aguilar (Chico State Alum, Master's from UCLA)
Title: Fuzzier Forests: Identifying Interactions
Abstract: Fuzzy forests are a computationally feasible feature selection method for correlated, high dimensional data. This presentation introduces new capabilities to fuzzy forests that allow researchers to identify interactions among features. In particular, these methods identify interactions within and across modules. Although random forest variable importance measures can account for interactions, users are unable to identify those interactions from the resulting output. Knowing which features are involved in interactions can elucidate the relationships of the features in the model. The performance of the across and within methods was evaluated using simulated data. The two methods can reliably select the interaction terms included in the true regression models. Additionally, an informal assessment for the interchangeability of correlated features provides evidence that, when using variable importance for feature selection as opposed to maximizing predictive accuracy, the correlated features do not act as proxies for one another.
Friday, Oct 8 -- Holt 171, 4:00pm
Nicholas Lytal (Chico State University)
Title: An Introduction to Single-Cell RNA Sequencing and Related Topics
Abstract: In the field of Bioinformatics, single-cell RNA-sequencing (scRNA-seq) is a subject of recent interest that explores differences among tissue samples on a cell-by-cell basis rather than as a whole. This process has applications in fields such as cancer research, embryonic studies, and more. However, sampling on such a scale has restrictions that often limit the power of scRNA-seq results.
In this talk, we will explore the basics behind scRNA-seq, as well as some of the statistical methods that have been developed to improve its results.