Gustavo Rohde is developing a new class of modeling tools

Today’s ready availability of data has not supplanted traditional hypothesis-driven methods of research, but it has created a more fluid interplay between the exploration of data and the formulation of hypotheses. Rather than progress in a linear fashion from observation to hypothesis, researchers might begin with a scientific challenge and gradually refine a hypothesis as they explore the data.

In other words, researchers in many fields don’t rely on data exclusively to prove a hypothesis. They also use it to help formulate one. And to do this, they need better tools to help them grasp the meaning of data.

Associate Professor Gustavo Rohde’s Imaging and Data Science Lab in the University of Virginia Department of Biomedical Engineering has responded to this shift by creating tools that help researchers more productively explore data. “Our lab is focused on creating computational methods that allow scientists to interact with and explore datasets in biology and medicine so that they can begin asking questions,” Rohde said.

In March, Rohde received a grant from the National Institutes of Health to devise a new class of modeling tools that are better adapted to exploring biomedical data than the tools currently available.

Open portrait of Gustavo Rohde

Gustavo Rohde, PhD, Associate Professor of Biomedical Engineering and Electrical and Computer Engineering

Tools That Emerge from the Mathematics of Biological Processes

Rohde works at the intersection of mathematics and modeling, using mathematics to describe natural phenomena and building tools based on these mathematical descriptions. In this grant, he addresses a fundamental mismatch between mathematics being used to create tools for biological processes and the biological processes themselves.

The datasets that biomedical researchers explore typically consist of data points describing an underlying biophysical phenomenon, such as the voltage of a heartbeat or the distribution of gray matter within the brain. In many cases, these datasets are in the form of a signal or an image, both Rohde lab specialties. A signal might be an EKG or a pulse. An image might be a CT scan or MR image.

Many physical properties that biomedical researchers would like to understand involve transport, or movement of some kind, whether it be a change in intensity or shape. The mathematics behind the algorithms currently used to model this transport, however, are not grounded in physical processes. For instance, researchers interested in exploring changes in the concentrations of a molecule in a cell can use existing modeling tools to track fluctuations in concentrations at specific locations. However, these tools—because they are uncoupled from cellular processes—do not reveal where molecules go as the concentration at specific locations moves up or down.

“We are starting with the mathematics of transport and building computational tools based on it that will give researchers a fuller understanding of the biomedical datasets they are exploring,” Rohde said.

Building an Entirely New Model from the Ground Up

As part of the grant, Rohde will develop the new mathematical modeling framework and create an open-source software package containing modeling components that other researchers can use as needed to analyze datasets. These mathematical descriptions will be based on transforms, or equivalencies, Rohde and his lab have developed to unite different ways of looking at signals or pixels in images. Preliminary application of these transforms has been shown to augment image classification accuracy in several clinical areas, including drug screening and differentiating cancerous from normal cells.

Rohde’s goal is to provide researchers with tools they can combine as necessary to rapidly prototype predictive models and formulate more accurate hypotheses.

“Our aim is to greatly enhance the ability to extract meaning from diverse biomedical datasets, while augmenting the accuracy of predictions,” he said.

His new framework may also be more computationally efficient and require less set up than existing approaches. Ultimately, these tools could accelerate the pace of biomedical discovery.