Events Calendar

Predictive Algorithms & Cyberinfrastructures for Precision Medicine

Somali Chaterji
Cells and Machines Innovatory
Purdue University

Abstract
Precision medicine emphasizes the way in which health and disease vary in every single individual or in cohorts of individuals. This is because the health of an individual is a function of her genome, epigenome, and metagenome. Every cell of the human body has the same genome. How then is a brain cell distinct from a heart cell? How does a cancer cell bypass the checks and balances of the immune system? This is where the cell’s epigenome offers a distinct “symphony” to diverse cellular states in varied contexts (level of maturity, pathogenicity, etc.). Driven by the exabytes of sequencing data being generated, there is an increasing need to analyze genomic big data to interpret and stratify disease for personalized therapies. How can this “genomical” big data enable the strides of precision medicine? What kinds of algorithms can deal with the inherent heterogeneity, the noise, and the high-dimensionality of this kind of data? Are there recurrent kernels or motifs in these algorithms that can be identified to speed up development of new algorithms? Can these efforts result in precise data-driven medicine?

In this talk, I will answer some of the above questions through our ML algorithmic suite— Avishkar. First, in our Avishkar suite, we uncover the non-canonical signatures of small regulatory RNA (e.g., microRNA, miRNA for short) targets. Regulatory miRNA-gene interactions are known to control a vast swath of cellular processes. Using our suite of predictive algorithms (SVM and ANN variants), we are able to predict miRNA targets in a disciplined manner. Then, I will present a few examples of context-aware interactions that can uncover combinatorial regulatory effects whereby multiple miRNAs together regulate expression of individual genes or clusters of genes dependent on the cellular context. Our modular and distributed implementations of the models are effective both from the standpoint of speed and for biological relevance, say for example varied gene regulatory interactions in a healthy person, distinct from a person in the throes of an anomaly. Finally, I will present our work on federated cyberinfrastructures for genomics. This is in the context of the MG-RAST repository, the largest metagenomics portal and analysis pipeline and hosted by the US Department of Energy.

Tuesday, November 27, 2018 at 11:00am to 11:50am

Learning Innovation Center (LINC), 268
165 SW Sackett Place, Corvallis, OR 97321

Event Type

Lecture or Presentation

Event Topic

Research

Organization
College of Engineering, Chemical, Biological, and Environmental Engineering, Electrical Engineering and Computer Science
Contact Name

Jonathan Rich

Contact Email

jonathan.rich@oregonstate.edu

Subscribe
Google Calendar iCal Outlook

Recent Activity