Jason Lu, PhD meets with colleagues.
Jason Lu, PhD (second from left) is an Associate Professor with the Division of Biomedical Informatics at Cincinnati Children’s Hospital and the UC Department of Pediatrics.

Pathogens, genetics, anatomy, environmental factors—how can we understand the complex mechanisms of human diseases? Jason Lu, PhD has dedicated his professional life to answering this question.

With thousands of known diseases and an infinite combination of factors, it is a big question to take on. But Lu has developed an effective strategy: Computational tools. His research lab at Cincinnati Children’s Hospital uses quantitative approaches in computer science and mathematics to unravel the molecular mechanisms of human diseases. Throughout his career, Lu’s skills in math and computer science have gone a long way in improving children’s health.

Exploring new approaches

As a graduate student at the Washington University School of Medicine (WUSM), Lu had a close encounter with the Human Genome Project. The year was 1998, and WUSM was one of the major contributors to sequence the entire human genome—everyone was excited about the possibilities. Lu could see that the anticipated discoveries would require a lot of computational expertise. But as a computational biology student, he was not expected to be able to resolve important biological problems by developing computational tools.

“At that time, people still thought of computational biology as just an aid to wet lab experiments,” says Lu. “But I thought of my computational skills as an alternative way to study biology. For every biology subdiscipline, there should be both experimental and computational approaches, and they should be equally important. They are two sides of the same coin.”

Technological breakthroughs had transformed many techniques into high-throughput—genome sequencing, x-ray crystallography—and Lu began to take part in these discoveries. As a graduate student, he studied the structure and function of proteins, developing an algorithm to predict protein-protein interactions based on shape and charge. This resulted in the first genome-scale structure-based protein interaction map in budding yeast.

Understanding human diseases

As Lu moved on to postdoctoral research at Yale University, the concept of network biology emerged. This new approach provided models for interpreting genome data in the context of health and disease.

“I started to think,” he says, “about how people traditionally study protein interactions on a smaller scale. But in the post-genomic era, we have the ability to look at interactions among thousands of proteins. Those interactions can be viewed as protein interaction networks. People have been studying different types of networks for decades, and many of these techniques can be borrowed to study protein and gene networks.”

Lu took part in a study to predict protein interaction networks based on evidence from genomic features, sequencing, protein structures, gene essentiality, and co-evolution of genes. The key idea was to look at different types of evidence to reach a final conclusion. By integrating protein structure information into protein networks, he and his colleagues also gained insight on the evolution of protein networks.

When Lu started his independent research lab at Cincinnati Children’s, he continued this line of research, but began to explore new areas. After learning that one of the most common types of data in the healthcare industry comes from medical imaging, he became interested in analyzing these images. His first priority was developing algorithms that could interpret the vast amount of medical imaging data. 

“In collaboration with the Imaging Research Center, we developed machine-learning algorithms based on pre-implantation brain imaging to predict the effectiveness of cochlear implants in children,” says Lu. “Cochlear implantation is one of the surgeries that hospitals use to help hearing impaired children improve speech ability, but it is not always effective. About 30 percent of children who receive cochlear implants do not develop the expected language development benefits. The surgery is also expensive and time-consuming. If we can determine which kids are likely or unlikely to gain the expected benefits of cochlear implantation, then we can consider other treatment options and strategies.”

While developing these algorithms, Lu also studied the classification of normal hearing from hearing impaired infants based on MRI images. Researchers were able to classify these groups with 95 percent accuracy, suggesting that brain imaging could be very useful in the diagnosis and treatment design of many diseases—especially neuropsychiatric diseases.

“In Autism Spectrum Disorder, for example, there is currently no objective method of diagnosis,” says Lu. “Most diagnoses are based on qualitative data such as Q&A and expert opinion. We developed a method to diagnose autism from MRI data using deep neural network with novel feature selection. The clinical significance is our ability to be objective, which could result in a paradigm shift in evaluating treatment options.”

Combining research perspectives

What are the factors that lead to a diagnosis of autism? The answers vary depending on research perspectives—biological, biomedical, and clinical—and Lu aims to combine them all. 

“We have to look at different types of large scale data in order to gain a complete perspective and understanding of autism,” says Lu, “and we are using this approach to study various diseases. My lab can analyze diverse types of data from genomics, imaging, environmental factors, and microbial studies. This gives us the unique ability to conduct integrative studies.”

Lu wants to ensure that his lab’s discoveries are not only theoretical or academic, but also relevant to clinical needs. His team is working with radiologists at Cincinnati Children’s to test the effectiveness of medical imaging algorithms in a clinical setting. Algorithms receive patients’ imaging data first, filtering information to give confident decision support. If the algorithm cannot make a confident decision, the data is then passed along to a radiologist who can decide whether the prediction is accurate.

“Testing and improving our algorithms dramatically reduces the workload of radiologists, improving the turnaround for diagnoses in children,” says Lu. “This is another way that we can more directly impact the clinical improvement of children’s health.”

Jason Lu, PhD is an Associate Professor with the Department of Biomedical Informatics at Cincinnati Children’s Hospital and the UC Department of Pediatrics. He is pursuing research in computer science, applied mathematics, and molecular mechanisms to study human diseases.

For more information, visit the Lu Laboratory of Bioinformatics and Systems Biology website, or contact long.lu@cchmc.org.