Pitt Computer Engineers Win Best Paper Award at International Big Data Conference
TUSCANY, ITALY (December 5, 2017) … A team of computer engineering and bioengineering researchers from the University of Pittsburgh won the Best Paper Award at the 3rd International Conference on Machine Learning, Optimization & big Data (MOD 2017). The paper titled “Recipes for Translating Big Data Machine Reading to Executable Cellular Signaling Models” describes how automated machine reading can be used to pore over volumes of research and use that information to create models for understanding biological processes.
“These models are used to conduct and explain hundreds of thousands of simulated experiments, which would be impractical if done with biological material in the lab,” says Natasa Miskov-Zivanov, assistant professor of electrical and computer engineering at Pitt’s Swanson School of Engineering. “Our paper won the Best Paper Award because the methods it presents are critical to automating the process of model generation from vast amounts of literature without human intervention.”
The MOD 2017 conference judges selected eight finalists for the Best Paper Award after calling on the international machine learning community for submissions. A major theme of the conference is developing ways to improve computation tools to automate tedious tasks like reading through thousands of studies to find the specific information that applies to a particular biological niche.
“Creating big models to describe biological processes is still mainly a manual effort,” explains Khaled Sayed, a PhD student in Dr. Miskov-Zivanov’s lab. “Natural language processing tools are getting better at extracting useful information, but there is still a disconnect between how the tools work and the how researchers assemble models. Our paper describes an interface we have built to connect the efforts of machine reading and model building in biology.”
Model building uses equations, data structures, and conceptual tools to represent biological systems in computer simulations. By doing so, researchers are able to better understand biological processes through testing and analysis. The more detailed the model, the more likely it can be used to design new disease treatments and guide future research. Dr. Miskov-Zivanov and her team developed a language processing tool so humans and machines can interpret big data more accurately.
“The interface could be applied immediately to existing literature. Several state-of-the-art reading engines are already using it for their output, and we have applied it to reading hundreds of thousands of papers from PubMed to build and expand models of pancreatic cancer and melanoma cells,” says Dr. Miskov-Zivanov.
In addition to Sayed and Dr. Miskov-Zivanov, the paper’s authors included: Adam Butchy, PhD student in the Department of Bioengineering, and Carnegie Mellon University’s Cheryl Telmer, research biologist at the Biological Sciences and the Molecular Biosensor and Imaging Center.
Dr. Miskov-Zivanov led the project and proposed the initial version of the representation format. Dr. Telmer worked with the engineers to explain biological information and provide insight on how it could be represented. Sayed analyzed the features of the reading output and worked to develop the most suitable representation format, and Butchy identified the times when data was extracted correctly and misinterpretations. He also presented the paper at the MOD conference.
Contact: Paul Kovach