In the age of big data for biomedicine, images and other digital representations of the body are more available than ever before. Yu-Ping Wang, professor of biomedical engineering at Tulane University, is taking advantage of this time to tackle some of medicine’s most complex diagnostic challenges.
Wang runs the Multiscale Bioimaging and Bioinformatics Laboratory, which aims to leverage artificial intelligence in developing computational tools that take as input very different data types and produce as output more accurate, robust predictions of health outcomes, mainly in psychiatric disorders.
“There are not too many powerful tools in machine learning or the AI field that can identify the complex relationship between different data types,” Wang said. “So, we [are pioneering] the work in multimodal … heterogeneous data analysis, to integrate complementary information at different scales.”
With appointments in the departments of biostatistics and bioinformatics, biomedical engineering, Wang’s research production is wide-ranging. Most recently, his work has focused on applying collaborative learning –– a subtype of machine learning –– to enable diverse forms of neuroimaging and genomic data to speak to each other, generating more richly informative results in studies of development and psychopathology.
Researchers initially developed collaborative, or federated, learning to improve data privacy, as it allows a central server to distribute a global model to local devices, which train it on their private data and send back only model updates, not the data itself. But collaborative learning architectures can also be applied for “multimodal fusion” by first training encoder models separately on each dataset, then aggregating their outputs.
In one of his most cited articles, Wang and collaborators found that incorporating group-level constraints about brain structure into the mathematical formulation of a correlation allows better recovery of the true associations between genetic mutations and the brain activity of people with schizophrenia. His group has since extended this work, using multimodal fusion methods to explore variations in brain structure and function between males and females over the course of adolescence.
Beyond his chosen methodologies, collaboration and multi-disciplinarity are central to Wang’s ethos as a researcher.
“I studied math, computer science and biomedical engineering. I also have industrial experience,” Wang said. “So, I also recruit students from different disciplines … so when we work together, we can make the most use of our skills.”
Last month, Wang received a Research Project grant, or R01 grant, from the National Institutes of Health –– nearly $2 million in federal funding –- to continue his study of the complex causes of mental illnesses like schizophrenia. After several years of struggle following the COVID-19 pandemic to secure lab funding and graduate students, the grant comes as a welcome relief.
“I really want to use this grant to expand the capacity of my lab, to bridge the gap between the new AI technology and brain imaging and multi-omics,” Wang said.