Abstract

Plants are essential components of our ecosystem and play a vital role in sustaining life on earth. Recent climate changes pose significant challenges to agricultural productivity by exposing crops to increasing biotic and abiotic stresses. Developing resilient crops requires an integrated understanding of plant development, genetics, and physiology, coupled with innovative technologies for crop improvement. This talk will highlight my research on plant genome engineering, peptide-mediated immune responses, root development, and stress tolerance. It will also introduce my current research investigating the genetic and developmental mechanisms regulating tillering in drought-tolerant, stay-green sorghum. By integrating molecular biology, functional genomics, and plant physiology, my research aims to identify strategies for improving crop traits and advancing sustainable agriculture under changing climatic conditions.

Speaker Bio

Dr. Sowmya Poosapati is a Research Specialist in Dr. Tesfamichael Kebrom’s laboratory at the Cooperative Agricultural Research Center (CARC), Prairie View A&M University (PVAMU), where she studies the genetic and developmental mechanisms underlying shoot branching in crops. Originally from Andhra Pradesh, India, she earned her Ph.D. in Biotechnology from Jawaharlal Nehru Technological University, India. Her doctoral research identified thermotolerant Trichoderma strains with enhanced biocontrol potential against plant pathogens, which were subsequently commercialized for agricultural use in India.

Dr. Poosapati spent over five years at DuPont India Ltd., receiving multiple awards for her contributions to genome editing and cloning projects. She was later selected as a Tata Institute of Genetics and Society (TIGS) Visiting Scholar, working with Professor Martin Yanofsky at the University of California, San Diego (UCSD), to develop genome-editing tools for rice.

She subsequently conducted postdoctoral research in Professor Alisa Huffaker’s laboratory at UCSD, investigating plant innate immune responses in Arabidopsis and legumes. At the Salk Institute for Biological Studies, she worked with Professor Joanne Chory to develop strategies for improving root architecture, particularly deeper and denser root systems, to enhance plant performance and contribute to climate change mitigation. Her research spans plant molecular biology, genome engineering, plant immunity, root development, and crop resilience.

Congratulations to Thanmayee Yeluri, a Ph.D. student in the Department of Electrical and Computer Engineering and Graduate Research Assistant at the Center for Computational Systems Biology (CCSB), whose recent research submission has been accepted for a Short Talk presentation at the 20th International Conference on Data and Text Mining in Biomedical Informatics (DTMBIO 2026)!

Thanmayee will travel to Osaka, Japan, to present her research, co-authored with her advisor Dr. Seungchan Kim. The conference will be held at the Osaka University Nakanoshima Center from October 6 to October 10, 2026.

Title: Multimodal Machine Learning with Cell-Specific Regulatory Features for Single-Cell DNA Methylation Prediction

Authors: Thanmayee Yeluri and Seungchan Kim

Abstract

Single-cell DNA methylation measurements are often sparse, making it difficult to characterize methylation patterns across cells and genomic regions. Multi-omic measurements from the same cells provide complementary information that may help predict methylation at unobserved cell–gene pairs. Here, we developed a multimodal machine-learning framework for cell–gene DNA methylation prediction using genomic sequence, RNA expression, chromatin accessibility, transcription factor (TF) features, CpG context, and regulatory annotations.

We compared XGBoost, CatBoost, a multi-layer perceptron (MLP), and a knowledge-guided neural network using a within-cell gene-split evaluation. Using the gene-associated feature representation, $x_{\cdot,g}$, the MLP and Knowledge-Guided models performed similarly, with Pearson correlations of 0.6066 and 0.6072, respectively. We then incorporated cell-specific RNA, accessibility, and TF-related features, $x_{c,\cdot}$, to construct the full multimodal feature representation, $x_{c,g}$. With this representation, the MLP reached a Pearson correlation of 0.6222, whereas the Knowledge-Guided model achieved 0.6523, with $\text{RMSE} = 1.2655$ and $R^2 = 0.4249$. The Knowledge-Guided model also improved cell-level correlation for 79.4% of evaluable cells and reduced prediction bias across the methylation range.

Ablation experiments showed that cell-specific TF expression contributed most strongly to the improvement associated with the cell-specific regulatory information, whereas ablating TF motif, TF-expression–motif interaction, or learned gating components did not reduce performance. These results show that cell-specific molecular information can improve within-cell methylation prediction and identify TF expression as the most informative cell-specific regulatory component in the current framework.

Abstract

Family health history (FHH) is a valuable and accessible genomic tool for identifying individuals at increased risk for cancer and guiding personalized prevention strategies. However, many underserved populations face barriers to collecting, understanding, and using FHH information for cancer prevention. This presentation will highlight a series of community-based participatory research projects focused on FHH-based cancer prevention among Chinese Americans in Texas. The presentation will describe the development, implementation, and evaluation of culturally and linguistically tailored interventions designed to improve cancer risk awareness, family communication, and engagement in preventive services. Particular emphasis will be placed on strategies that train and mobilize community health workers (CHWs) to deliver FHH-based cancer education, risk assessment, and navigation services within their communities. Findings from colorectal, breast, and other cancer prevention initiatives will be discussed, demonstrating how community-engaged approaches can increase access to evidence-based prevention services and reduce cancer disparities in underserved populations.

Speaker Bio

Dr. Yulyu Yeh is an Assistant Professor in the Division of Health, School of Public and Allied Health at Prairie View A&M University. She earned her Ph.D. in Nutrition and Food Science from Wayne State University. Her research focuses on cancer prevention, family health history, nutrition education, and culturally tailored public health programs for underserved populations. Prior to joining Prairie View A&M University, she served as an Assistant Professor and International Graduate Student Program Director in the Department of Healthcare Administration at Asia University in Taiwan. She also completed postdoctoral training at Texas A&M University, where she managed multiple community-based research projects.

Dr. Shuvo was recently awarded a PVAMU RISE Undergraduate Award, “Toward Trustworthy and End-to-End AI for Next-Generation Computational Therapeutics”, to develop a trustworthy, explainable, and secure end-to-end AI framework that integrates multiple computational therapeutics tasks—from protein function prediction to drug response prediction—to accelerate and improve drug discovery. This is a 1-year grant to support an undergraduate student, up to $7,200.

Project Title: Toward Trustworthy and End-to-End AI for Next-Generation Computational Therapeutics

Principal Investigator: Md Hossain Shuvo, Ph.D.

Project Description:

Drug discovery is a long and expensive process, with many potential drug candidates failing before reaching clinical use. Artificial intelligence can help reduce this burden by identifying and prioritizing promising drug candidates earlier, but most existing AI tools focus on individual tasks and do not work together as a connected system. This project will develop a modular, end-to-end AI framework that connects important computational therapeutics tasks, including protein function prediction, drug–target affinity prediction, toxicity prediction, and drug response prediction. The project will also incorporate self-assessment and explainability so that the system can provide confidence estimates, identify important features behind its predictions, and better understand when and why models fail. In addition, the project will examine the effects of low-quality and potentially adversarial inputs and develop a secure deployment prototype with basic safeguards such as input validation, access control, secure communication, and activity logging. The expected outcome is a working prototype that demonstrates how multiple AI models can be connected into a more reliable, transparent, and secure computational therapeutics workflow. The results will provide a foundation for future research on trustworthy AI for biomedical applications and support the development of larger NIH, NSF, and DoD proposals.

Sponsor: R&I Division at Prairie View A&M University Funding Amount: $7,200 Project Period: FY 2026–2027

Dr. Shuvo was recently awarded an NSF Research Initiation Award: Bridging the Gap between Sequence and Function with Ethical and Explainable AI for Protein Function Prediction. This is a 3-year, $445,686 grant, sponsored by DUE (Division Of Undergraduate Education).

Project Title: Bridging the Gap between Sequence and Function with Ethical and Explainable AI for Protein Function Prediction

Principal Investigator: Md Hossain Shuvo, Ph.D.

Project Description:

This project will develop advanced deep learning frameworks for protein function prediction by integrating protein sequence, three-dimensional structural, protein–protein interaction, and protein–nucleic acid interaction information within ethical and explainable AI models.

The research will explore transformer models, graph neural networks, convolutional neural networks, and hybrid learning approaches to improve the prediction of biological functions from protein sequences. The project will emphasize explainability, transparency, and robustness and will evaluate the developed methods using publicly available benchmark datasets and the Critical Assessment of Functional Annotation (CAFA) Challenge.

The project will also provide open-source software, web-based computational tools, and educational materials to support the scientific community, while expanding student research and workforce development opportunities.

NSF Award Details