The Rose RNAseq project is an ongoing venture carried out by Dr. Anthony Varghese in collaboration with faculty from the UWRF Plant and Earth Sciences Department. It is centered around a dioecious species of rose, R. Setigera, that is native to North America. R. Setigera gene expression has been studied by isolating mRNA and sequencing it using the Oxford Nanopore third generation sequencing platform. 

Participating educators and their students will be able to see how biological sequence data is obtained and be able to compare sequences in existing databases to establish the function of these genes. Teachers will have an opportunity to use this research project to develop modules involving Python, bioinformatics, and performing sequence comparisons using the UWEC Blugold Center for High Performance Computing.

A number of labs at UWRF use sequence analysis to determine the identities of genes expressed in various species of plants like Rose and Ninebark. A typical RNAseq run generates sequence data from thousands of RNA molecules ranging from 1 to 10s of Kbases. A key step in using this data is the functional genomics analysis: the use of computational techniques to identify the function of genes represented in the sequence data. While some aspects of function prediction based on sequence information is possible one at time using publicly available servers at the National Center for Biotechnology Information (NCBI), the size of data generated by sequencing experiments makes this approach impractical. As a result, the NCBI actively promotes setting up cloud-based solutions like ElasticBLAST (Camacho 2023).  

Dr. Varghese has been using these cloud computational techniques in his research for the last decade in addition to training undergraduate and graduate students in their use. Participating educators will be able to use this research project to develop modules involving Jupyter notebooks, bioinformatics, and Platform as a Service (PaaS) cloud computing projects. Their students will have an opportunity to learn about how cloud computing techniques work and will be able to apply this important modern technology to solve problems in their own communities.

At UWRF, Dr. Varghese is actively involved with simulating electrical activity in cells of the heart (Varghese 2016). Each cell is a system of about 30 nonlinear ordinary differential equations (Varghese 1997). With no closed form or analytical solutions available for such stiff nonlinear equations, the only way of studying such systems is to solve them numerically using implicit ODE solvers (Varghese 1997). Besides evaluating numerical methods in Python or Java, teachers and their students can run simulations to study normal physiology and pathophysiology (Noble 1998, Varghese 2015) and learn about managing the data generated by these simulations. 

Participating educators can use this research project to develop modules involving numerical methods for time integration of differential equations and stability analysis using Python or Java. These modules can be introduced to students at the high school level as well.

Ion channels are proteins present in almost every cell of the body. They regulate the movement of ions between the inside and outside of cells to generate electrical activity such as action potentials in nerve and heart cells. Dr. Varghese has been studying (Varghese 2002, Piper 2003) a particular ion channel called the human Ether-a-gogo Related Gene (hERG). Ample experimental data is available from collaborators in Sydney, Australia (Vandenberg 2006) and the goal is to come up with Markov state models that can explain the available experimental data. Many such models have been built incrementally, and it is getting difficult to fit models to all available data. Participating educators can use this research project to learn how to fit mathematical models to experimental data using regression analysis and explore techniques such as particle swarm optimization.

DNA methylation is a process that affects gene accessibility and therefore gene expression. It involves adding a methyl group to the Cytosines in the DNA of a living organism (Moore, 2013). There has been a research focus on investigating methylation information to predict the relationship between specific genes and cancer. In this project, teachers will examine the potential of developing a scalable deep-learning framework that can process high-dimensional genomic datasets and identify methylation sites in the human genome responsible for tumors. The objectives of this project include implementing a dynamic feature selection algorithm capable of removing irrelevant features with quantifiable confidence and support.

Techniques such as ANOVA F-test, and Random Forest will be used to retrieve a list of notable features. This will be followed by implementing a fully connected deep neural network to develop a diagnostic prediction model for cancer detection from CpG methylation states. This project will begin with an introduction to machine learning for two weeks followed by the fundamentals of deep learning for a subsequent two weeks. Data from the Cancer Genome Atlas (Tomczak 2015) will be used to construct the prediction model. The high dimensionality of these datasets with each sample having up to 485,000 methylation sites, data will require processing using the Blugold Center for High Performance Computing (BCHPC) at UWEC.

Cybersecurity events are happening more frequently, therefore it's important to keep up with the constantly changing dynamics of malicious internet traffic. Although these internet packets can be easily detected using traditional machine learning methods, the underlying threat detection framework requires ongoing retraining which can be expensive. The issue is made more challenging because datasets are highly unbalanced for malicious internet traffic. It is also important to note that most of these solutions require real time response which is challenging on legacy systems due to issues such as the large dataset dimensionality. One of the most effective ways to reduce computational complexity is to manually reduce feature size. Keeping the essential components while eliminating the unnecessary ones can increase model’s predictive performance and prevent overfitting. This manual feature reduction problem can be addressed by a Distributed Feature Selection (DFS) strategy. Participating teachers will explore DFS methods that combine Univariate Feature Selection, Correlated Feature Elimination, Gradient Boosting, and Wrapper Methods to automate crucial feature selection. To predict harmful traffic, the results are channeled into a pipeline made up of different machine learning algorithms. Multiple cybersecurity datasets will be used to test the proposed approach such as NSL-KDD and UNSW-NB15. The first two weeks will be an introduction to machine learning algorithms used in this study while the remaining two weeks will incorporate building the DFS strategy, training the proposed architecture, and hyperparameter tuning.

The 5-year overall survival rate for Pancreatic Ductal Adenocarcinoma (PDAC), an aggressive abdominal cancer, is around 6% (Noda 2022). The number of cases and fatalities in the United States is expected to climb by a factor of two in the coming decade and will account for more than 90% of all related pancreatic cancers. Hence, this project by Dr. Gomes aims to explore state-of-the-art imaging techniques to diagnose this disease.In stage 1 of this project, the participating teachers will explore the concepts of image processing such as normalization techniques and filters that can be used to extract meaningful information from data. Teachers will also be able to process CT scan formats such as DICOM. In the following two weeks, they will explore deep learning approaches such as Convolutional Neural Networks (CNN) and assess its potential to process image datasets. They will create a UNet-based CNN architecture from scratch with the goal of increasing prediction accuracy. This UNet can then be applied to portal venous phase CT scan slices from public datasets available through the Medical Segmentation Decathlon (Antonelli 2022) and the Cancer Imaging Archive (Clark 2013). The CNN architecture will be implemented on the GPU nodes of the BCHPC at UWEC. Participants in this project will experience the computational aspects of deep learning along with the application of these computing concepts to advance patient healthcare.