Title: DART Seed Grant Recipient Presentations, Session 2
Presenter(s): Yasir Rahmatallah, Jason Causey, Sudeep Bhattacharyya, and Melody Greer
Date: October 26, 2022

Description

This month’s webinar will feature three presentations from the recently awarded DART YR 3 Seed Grant Recipients.

  • Yasir Rahmatallah (UAMS) will present on Machine Learning Approaches for Remote Pathological Speech Assessment for Parkinson’s Disease. This project is part of the LP research theme.
  • Jason Causey (A-State) will present on MoDaCoM-TL: Model and Data Compatibility Metric for Transfer Learning. This project is part of the LP research theme.
  • Sudeep Bhattacharyya (A-State) and Melody Greer (UAMS) will present on Smart curation and deep learning-based enhancement of social risk data. This project is part of the DC and LP research themes.

Machine Learning Approaches for Remote Pathological Speech Assessment for Parkinson’s Disease (Rahmatallah)

The objective of this project is to develop analysis pipelines to detect discriminative speech acoustic features between people with Parkinson’s Disease and healthy controls using voice recordings acquired via phone line. Parkinson’s disease is a neurodegenerative disease of aging that leads to a spectrum of symptoms including speech impairments. Speech impairments usually manifest in the form of decreasing speech intensity and speed, monotone voice, imprecise articulation, change of pitch, tremors, hoarseness, breathiness, and freezing of speech. Diagnostic criteria are based on clinical motor examination, and motor symptoms manifest over years and possibly decades after the pathology starts. Diagnosis may be delayed for millions of patients living in rural under-served areas with limited access to neurological care. The project will harness the predictive power of machine learning methods while increasing the interpretability of the predictions by highlighting the specific impairment speech patterns associated with the most discriminative features. The proposed project will provide a much needed foundation for validation and methodology development studies and for distinguishing speech patterns characteristic of specific phenotypes of Parkinson’s patients.

MoDaCoM-TL: Model and Data Compatibility Metric for Transfer Learning (Causey)

Training modern deep learning models from scratch requires huge datasets and large amounts of computational time, both of which are difficult to justify for scientists applying these models to real-world questions.  Transfer learning, in which a model that is already trained on data similar to the target is altered to allow predictions in the new domain, can provide an efficient mechanism to allow practitioners to utilize these deep models. However, not every pre-trained model performs well when transferred to a new problem domain.  Likewise, pre-training new models on existing datasets requires matching existing data to the new target for a successful knowledge transfer.  This project aims to address these two challenges by providing metrics to help researchers match datasets to model architectures and targets.

Smart curation and deep learning-based enhancement of social risk data (Bhattacharyya and Greer)

Socioeconomic and behavioral aspects of our lives significantly impact our health, yet minimal social determinants of health (SDOH) data is collected in the healthcare system. Information of this type is needed for quality healthcare research and patient care because it is associated with the full spectrum of health outcomes, from acute to chronic disorders. Despite the growing interest in including SDOH data in the healthcare process, there are significant challenges associated with its collection, management, quality, and portability. We are working to address the need for standardized and complete SDOH information by developing a fully automated SDOH data collection and curation process at the individual patient level and create deep learning-based imputation models to produce complete, high quality data.

Presenter Bios

Dr. Yasir Rahmatallah is an assistant professor in the Department of Biomedical Informatics at the University of Arkansas for Medical Sciences. He received his PhD degree in Applied Sciences from the University of Arkansas Little Rock in 2010. His research interests include Bioinformatics and signal processing, with focus on the development of statistical methodology for gene set analysis of large-scale omics datasets. He is the developer and maintainer of the Bioconductor package GSAR which provides statistical methods to detect gene sets with specific alternative hypotheses. The package has been downloaded over 18,000 times between 2015 and 2021 with an average of 2,500 downloads per year. His recent research projects include the secondary use of electronic health records to predict disease development trajectories and treatment factors influencing them, and using machine learning approaches for remote pathological speech assessment for Parkinson’s disease patients. Dr. Rahmatallah supports research in primarily undergraduate institutions in Arkansas as part of the Arkansas INBRE research support core.

Dr. Jason L. Causey is Assistant Professor of Bioinformatics in the Department of Computer Science at Arkansas State University. He received his Ph.D. from the UALR / UAMS Joint Graduate Program in Bioinformatics at the University of Arkansas at Little Rock, where his work studied low-complexity structures in biological and biomedical data. His research focuses on mathematical and computational approaches to scientific questions, particularly for biomedical imaging and genomic studies. Dr. Causey serves as Associate Director of the Center for No-Boundary Thinking (CNBT), and as the Division Lead of the CNBT Division of Algorithms and Computational Methodology. Dr. Causey also serves as Associate Director of the Arkansas State University / St. Bernards Translational Research Lab (TRL).

Dr. Sudeep Bhattacharyya is an associate professor of Bioinformatics and Data Science at Arkansas State University, Jonesboro. Prior to moving to Astate, she was a tenured associate professor in the department of Biomedical Informatics at UAMS. Her research interest is twofold 1) gaining mechanistic insights into neurodegenerative diseases and response to therapy using data-driven modeling and systems biology approach and 2) application of AI in health analytics, combining patient level and population level data for better patient care and health outcomes. She serves in the leadership committee of the Mental Health section of American Public Health Association, is the program chair-elect of Statistical Learning and Data Science section of American Statistical Association and is a member of the Justice, Equity, Diversity and Inclusion Task Force of the American Statistical Association. She has served in leadership roles at ASCPT.org, served in think tank sessions at NIH, review panels at NSF and DHHS’s Office of Adolescent Health and serves in the editorial board of two reputed journals.

Dr. Melody Greer is an assistant professor in the department of biomedical informatics in the College of Medicine at the University of Arkansas for Medical Sciences. Her research is focused on social determinants of health data enhancement and interoperability in healthcare, as well as clinical trial data management. She is a faculty member for the UAMS Clinical Informatics Fellowship and a mentor for Health Sciences Innovation and Entrepreneurship postdoctoral fellows. Dr. Greer works to collect and manage clinical trial data as part of the Data Coordinating Center of the ECHO IDeA States Pediatric Clinical Trials Network. She also serves on the Translational Research Institutes Team Science Committee.