Presenter Profile
Mehmet Koyutürk
Professor
Case Western Reserve University, Department of Computer and Data Sciences
Mehmet Koyutürk is the Andrew R. Jennings Professor of Computing Sciences at the Department of Computer and Data Sciences at Case Western Reserve University (CWRU). He also holds a secondary appointment at the Center for Proteomics and Bioinformatics at the School Medicine, and serves in the steering committee of CWRU's graduate programs in Systems Biology and Bioinformatics. Mehmet received his Ph.D. degree (2006) in Computer Science from Purdue University. Mehmet’s research focuses on data mining and computational biology, with particular emphasis on the analysis of biological networks, systems biology of complex diseases, and computational genomics. To date, he has published more than 100 papers on these topics. His research is funded by multiple R01 research grants from US National In- stitutes of Health, as well as several grants from US National Science Foundation, including a CAREER award. Mehmet currently serves an associate editor for IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB).
TALK TITLE
Enhancing network biology algorithms to rigorously illuminate understudied proteins
KEYWORDS
Network biology, machine learning, understudied proteins
ABSTRACT
According to the Understudied Protein Initiative, “95% of all life science publications focus on a group of 5,000 particularly well-studied human proteins". Despite the explosion in the development and application of machine learning algorithms to drive knowledge discovery in systems biology, network biology algorithms offer very little insight into understudied proteins. In this talk, we argue that computational algorithms are not able to shed light into understudied proteins because they are not explicitly designed or evaluated for that purpose. In contrast, we demonstrate that predictive models in systems biology are validated using benchmarks that are biased toward well-studied proteins. As a result, algorithmic bias toward well-studied proteins is reinforced by the evaluation process, thus most predictive models that are available are fine-tuned to discover what is already known. Motivated by these observations, we develop algorithms to mitigate bias and enhance the reach of predictive models to understudied proteins and support these algorithms with evaluation strategies that take fairness into account. Our results on kinase activity inference, kinase-substrate association prediction, and protein-protein interaction prediction show that bias-aware network algorithms can be more robust to missing information and reliably discover knowledge on understudied proteins.