PNNL Collaboration Develops Clinical Predictive AI from TransMED

August 2, 2022 — The onset of the COVID-19 pandemic posed a huge challenge for health professionals. Doctors struggled to predict how different patients would fare under treatment against the new SARS-CoV-2 virus. Deciding how to classify medical resources when presented with very little information took a mental and physical toll on health care providers as the pandemic progressed.

To alleviate this burden, researchers from Pacific Northwest National Laboratory (PNNL), Stanford University, Virginia Tech and John Snow Labs have developed TransMED, a first of its kind artificial intelligence (AI) prediction tool aimed at addressing problems caused by emerging or rare diseases.

“As COVID-19 unfolded in 2020, it brought some of us together to think about how and where we could make a meaningful contribution,” said lead scientist Sutanay Choudhury. “We decided that we could have the most impact if we worked on the problem of predicting patient outcomes.”

“COVID posed a unique challenge,” said Khushbu Agarwal, lead author of the study published in Nature Scientific Reports. “We had very limited patient data for training an AI model that could learn the complex patterns underlying COVID patient trajectories.”

The multi-institutional team developed TransMED to address this challenge by analyzing data from existing diseases to predict the outcomes of an emerging disease.

From left to right, PNNL researchers Robert Rallo, Sutanay Choudhury, Khushbu Agarwal and Colby Ham contributed to the development of TransMED. Credit: Andrea Starr, PNNL.

Answer a call for help

When the COVID-19 pandemic started, PNNL researchers took on the new challenge right away. Choudhury found himself working on a team that used AI to generate structures for molecules that could be potential candidates for drug development against SARS-CoV-2.

He also felt an intense empathy for the health professionals on the front lines of the COVID-19 battle. “It was clear that we needed to build more effective tools to better protect both patients and caregivers during the next crisis,” Choudhury said.

Choudhury and Agarwal enlisted the help of Colby Ham and Robert Rallo, director of the Advanced Computing, Mathematics, and Data Division at PNNL, as well as computer scientists from Stanford University, Virginia Tech, and John Snow Labs to build such a tool.

Suzanne Tamang was one of those scientists. She previously worked with Choudhury, Agarwal and Rallo on a healthcare analysis project. She was eager to participate in this study in order to apply her knowledge to provide decision support to health professionals.

“We all saw the need to contribute,” said Tamang, assistant faculty director, Data Science, at the Stanford Center for Population Health Science and instructor in the Department of Biomedical Data Science, Stanford University School of Medicine. “We could use our capabilities to build a tool with immediate value and usability for health professionals.”

Tamang is no stranger to such altruism. As part of Stanford University’s Statistics for Social Good club, she regularly devotes her time and skills to problem-solving on various social issues. “Sometimes the best science comes when researchers are driven by the desire to help,” Tamang says.

A new approach to fighting unknown diseases

Early results indicate that TransMED outperforms current patient outcome prediction models, especially for rarer outcomes. Agarwal attributes this in part to TransMED’s ability to explore a wide range of medical information, including other respiratory diseases.

“TransMED takes into account almost all types of electronic health records, such as medical conditions, medications, procedures, laboratory measurements and information from clinical notes,” said Agarwal. “By taking this holistic view of the patient, TransMED can make predictions in the same way a doctor would.”

The other factor contributing to the success of TransMED is transfer learning. Essentially, transfer learning works by making a machine learning model work to solve a problem where there is a lot of data. The model then transfers this knowledge to solving similar problems. In the case of TransMED, researchers trained the model on known outcomes from patients with severe respiratory disease and applied that knowledge to predicting COVID-19 outcomes.

“Given a patient’s recent medical history, TransMED can predict a patient’s need for ventilators or other rare outcomes 5 to 7 days into the future,” Choudhury said.

Application of AI in healthcare practice is still in its infancy, but this work is a promising first step towards building a useful model for predicting patient outcomes. While TransMED has yet to be tested in a clinical setting, it offers an encouraging glimpse into the future of healthcare.

Other authors of this article are Sindhu Tipirneni and Chandan K Reddy of Virginia Tech; Pritam Mukherjee, Matthew Baker, Siyi Tang and Olivier Gevaert of Stanford University; and Veysel Kocaman of John Snow Labs. This work was supported by a PNNL Laboratory Directed Research and Development program.

About PNNL

Pacific Northwest National Laboratory leverages its distinctive strengths in chemistry, earth science, biology, and data science to advance scientific knowledge and address sustainable energy and national security challenges. Founded in 1965, PNNL is administered by Battelle for the Department of Energy’s Office of Science, which is the largest proponent of basic research in the natural sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit For more information about PNNL, visit the PNNL news center.

Source: Sarah Wong, PNNL

Leave a Comment