Read more about the awarded projects for the 2025 Propelling Original Data Science (PODS) Grants.
Track 1: Data science and AI methodology and applications
The As-If Machine (AIM): A Multi-Agent RAG System for Reducing Psychological Distance Through Personalized Narrative Simulations
Ceren Budak (School of Information) and Stephanie Preston (College of Literature, Science and the Arts)
The As-If Machine (AIM) is an interactive storytelling tool that helps people imagine their lives in different ”what if” scenarios. It uses a team of AI agents to create immersive personalized stories inspired by psychological theories on how narratives shape our thinking and emotions. PODS funding will allow us to test the impact of AIM on attitudes related to long-term risks, social divides, and trust in science and will allow us to develop AIM into a production-ready web application, which can be used by researchers in different fields.
SAGE: A Scalable GeoAI Framework for Zero-Shot Mapping of Lithium Mines
Joshua Newell (School for Environment and Sustainability) and Paramveer Dhillon (School of Information)
Demand for lithium—a critical mineral for electrification and decarbonization—is rapidly increasing, but we lack a clear, spatially explicit understanding of where and how it is being mined. Leveraging recent advances in geospatial artificial intelligence (GeoAI), we will build SAGE (Scalable AI for Geospatial Extraction)—a segmentation framework that combines remote sensing science with Meta’s Segment Anything Model (SAM), a state-of-the-art vision foundation model. This project will produce the first high-resolution global map and geodatabase of lithium mines and advance environmental monitoring, supply chain transparency, and decision-making in a vital priority area for national security.
Harnessing Evolutionary Legacies in Protein Space: Evolvability as a new target for molecular optimization
Luis Zaman (College of Literature, Science and the Arts)
This project explores how some proteins are better at evolving than others and asks whether this adaptability—or evolvability—can be intentionally built into or tuned out of newly designed proteins. By using AI tools trained on naturally evolved proteins, our team maps how mutations affect protein function and adaptability, focusing first on fluorescent proteins that are easy to mutate and test. The goal is to improve protein design methods so that scientists can build molecules that are not just functional today, but capable (or not) of adapting to future challenges.
ML-Powered Anomaly Detection at the Speed of Light: Algorithms and Applications for Secure Power Grid Operations
Shubhanshu Shekhar (College of Engineering) and Vladimir Dvorkin (College of Engineering)
Power grids form the backbone of national and local economies, yet they are increasingly exposed to anomalies such as extreme weather events, failures of aging infrastructure, and cyberattacks. This project brings together experts in statistical machine learning and power grid engineering with the long-term goal of developing rigorous algorithms for fastest anomaly detection in power systems with guaranteed speed and identification accuracy. To achieve this goal, the project proposes a cohesive fusion of classic quickest change detection theory, recent advances in predictive and generative machine learning, and modern power grid engineering.
Harnessing AI for Advancing Data Collection and Population-Scale Causal Inference
William Axinn (Ford School of Public Policy), David Jurgens (School of Information), and James Wagner (Institute for Social Research)
This project is designed to revolutionize general population causal inference by simultaneously accounting for all components of data creation errors while calculating causal associations. This new approach has the potential to advance U-M’s position as the leaders in data collection science as AI increases the breadth of data creation errors we can measure and address. Ultimately, the findings have the potential to drive both improvements in data quality and causal inference.
Track 2: Accelerating responsible AI research ecosystems
Facilitating Appropriate Reliance on Generative AI (GenAI) Tools by Investigating Reliance Decisions and Norms
Q. Vera Liao (College of Engineering)
This project will empirically develop and validate: 1) a framework to reason about the context-specific norms around appropriate reliance on GenAI tools to support defining and measuring appropriate reliance; 2) a framework outlining factors and their contributions to users’ reliance decisions to inform approaches that identify inappropriate reliance to mitigate the negative impacts.
AI Systems to Combat Non-Consensual Intimate Media (NCIM)
Sarita Schoenebeck (School of Information) and Eric Gilbert (School of Information)
Non-consensual intimate content, including sexual deepfakes and “revenge porn”, is a large-scale and growing societal crisis. Victim-survivors face relentless violations of their privacy and dignity and must manually report content across hundreds of websites, often with little success. This project builds web-based AI agents that act on behalf of victim-survivors to locate, report, and monitor non-consensual content across the web.
Governing AI’s Footprint: A Scalable Human-AI Workflow to Extract Zoning Codes for Data Centers and Renewable Energy Sitting
Xiaofan Liang (Taubman College of Architecture and Urban Planning) and Sarah Mills (Taubman College of Architecture and Urban Planning)
This project addresses the societal and environmental impacts of generative AI by targeting the opaque and inconsistent zoning regulations that govern data centers and renewable energy facilities. It proposes a human-AI workflow to extract structured zoning information across six Midwest states, creating a searchable, transparent, and replicable database. The resulting tools aim to support sustainable, equitable, and policy-aligned infrastructure planning in the AI era.
Track 3A: AI innovations in Clinical and Translational Sciences (CTS) (jointly funded by MICHR and MIDAS)
Developing Best Practices for AI-assisted Mixed Methods Analysis
Timothy Guetterman (Michigan Medicine) and Melissa DeJonckheere (Michigan Medicine)
This project will test and try to understand the quality and value of artificial intelligence for analyzing text-based data and for examining patterns, such as how processes differ between higher vs. lower performing healthcare systems. Traditionally, this analysis is computer assisted, but AI has the potential to save time and resources while maintaining quality.
Track 3B: AI impact and governance for health policy and healthcare (jointly funded by IHPI and MIDAS)
Implementing AI into Anticoagulation Clinical Decision Support
Geoffrey Barnes (Michigan Medicine), Michael Sjoding (Michigan Medicine), and Michael Lanham (Michigan Medicine)
Numerous AI tools have been developed, but few are currently used in routine clinical practice. We will compare the ability of machine learning and large language models to identify which patients with potentially inappropriate medication prescriptions are most likely to be recommended for a prescription change by clinical pharmacists. The findings from this project will be incorporated into a prospective trial evaluating the implementation and outcomes of an AI-informed clinical tool to improve safe medication prescribing embedded within the electronic health record.
Reducing Unplanned Hospital Readmissions with Causal Machine Learning
Jenna Wiens (College of Engineering) and Vikas Parekh (Michigan Medicine)
Hospital readmission rates are a publicly reported hospital quality measure; hospitals use transitional care interventions, like follow-up phone calls, to lower readmission rates. However, these interventions are often resource-intensive, and current methods for deciding who should receive them may not be effective. This study aims to analyze which patients actually benefit from these interventions using advanced machine learning with the goal of redesigning the resource allocation strategy to more effectively prevent readmissions, potentially influencing future policies at Michigan Medicine.
Track 3C: Data science and AI for health science and healthcare research
Enhancing Drug Combination Therapies Through Heterophilic Link Prediction with Graph Neural Networks
Danai Koutra (College of Engineering) and Sriram Chandrasekaran (College of Engineering, Michigan Medicine)
Drug combinations are increasingly used to combat drug resistance in cancer and infections, but the current trial-and-error approach to choosing them often leads to ineffective or unsafe combinations that can worsen clinical outcomes. While several computational models have been developed to predict optimal drug combinations, they often assume homophily (i.e., similarity in drug-target and drug-drug interactions) and fail to capture the heterophilous nature of many biological relationships wherein nodes frequently connect to dissimilar nodes. This project proposes a paradigm shift by characterizing heterophily in drug-protein interaction networks and interrelated biological networks, and introducing a novel heterophily-aware graph neural network (GNN) that jointly models diverse biochemical graphs to improve synergy prediction and enable interpretable drug combination discovery.
Predictive Modeling and Feature Learning for Large-Scale Neuroimaging Data
Jian Kang (School of Public Health) and Chandra Sripada (Michigan Medicine, College of Literature, Science and the Arts)
This project develops new artificial intelligence tools to better understand how brain development relates to learning, sleep, and life experiences in children and adolescents. Using data from the nationwide ABCD Study, we will create open-source software to predict developmental outcomes from brain scans while ensuring results are accurate, interpretable, and reproducible. The tools we build will help researchers and clinicians identify early signs of cognitive or health challenges and support timely, personalized interventions.
Revolutionizing Disease Diagnostics Through the Integration of Physics-Informed Materials Science Methods with Sequence Models
Sharon Glotzer (College of Engineering)
This project will develop an AI powered Python toolkit that will revolutionize the analysis of histology samples, by introducing quantitative, physics-informed metrics developed in the domain of materials science to histological analysis. Our toolkit boosts diagnostic accuracy by integrating spatially-resolved, single cell information into novel, sequence-aware machine learning models, circumventing the ubiquitous dearth of data that traditionally limits disease diagnosis from histology samples.