PredictMod: A Platform for Predicting Medical Intervention Outcomes and Sharing Custom ML/AI Models
Lori Krammer, Patrick M. McNeely, Urnisha Bhuiyan, Stephanie S. Singleton, Nikhil Arethiya, Abel Argaw, Vinod Aggarwal, Atin Basuchoudhary, Miguel Mazumder, John A. David, Sangeeta Agrawal, Sabyasachi Sen, Raja Mazumder · Network and Systems Medicine · 2025
Machine learning/artificial intelligence (ML/AI)-based predictive modeling is becoming increasingly popular in healthcare, particularly within the scope of precision medicine. Models based on patient data are effective tools for clinical decision-making, as well as impactful methods for researchers to expand on existing scientific and clinical understanding of biomarkers and biological pathways. We present PredictMod, an open-license platform for users to query, run, and publish electronic medical records and omics-based intervention outcome (responder vs non-responder) prediction models. The platform currently hosts several intervention outcome prediction models that utilize varied data types, conditions, and interventions, offering users insight into the kinds of information required to develop such models. Model-specific graphs and charts such as confusion matrices and feature importance facilitate user understanding of model output. Our model creation pipeline ensures comprehensive documentation using BioCompute Objects. The platform allows researchers in conjunction with clinicians to generate models using data from publications or their datasets and apply them to predict intervention outcomes for new patients or samples. PredictMod is particularly suited for small datasets, which are common in biological research. By providing a framework for pilot studies, the platform facilitates early-stage insights, helping researchers justify larger-scale data collection efforts. With its versatility and accessibility, PredictMod can advance biomedical research and precision clinical medicine by lowering technical barriers and fostering data-driven discovery approaches at the early stages of hypothesis generation using ML/AI.