In-Situ Data Analysis of Protein-folding Trajectories

Title: In-Situ Data Analysis of Protein-folding Trajectories

Speaker: Michela Taufer, University of Delware

Abstract: The transition from petascale to exascale computers is characterised by substantial changes in the computer architectures and technologies. The research community relying on computational simulations is being forced to revisit the algorithms for data generation and analysis due to various concerns, such as higher degrees of concurrency, deeper memory hierarchies, substantial I/O and communication constraints.

Simulations today typically save all data to analyse later. Simulations at the exascale will require us to analyse data as it is generated and save only what is really needed for analysis, which must be performed predominately in situ, i.e., executed sufficiently fast locally, limiting memory and disk usage, and avoiding the need to move large data across nodes.

In this talk, we present a distributed method that enables in-situ data analysis for large protein-folding trajectory datasets. Traditional trajectory analysis methods currently follow a centralized approach that moves the trajectory datasets to a centralized node and processes the data only after simulations have been completed. Our method, on the other hand, captures conformational information in-situ using local data only while reducing the storage space needed for the part of the trajectory under consideration. This method processes the input trajectory data in one pass, breaks from the centralised approach of traditional analysis, avoids the movement of trajectory data, and still builds the global knowledge on the formation of individual alpha helixes or beta sheets as trajectory frames are generated.

Short Bio: 

Michela Taufer is the David L. and Beverly J.C. Mills Chair of Computer and Information Sciences and an Associate P{rofessor in the same department at the University of Delaware. She earned her master’s degrees in Computer Engineering from the University of Padova (Italy) and her doctoral degree in Computer Science from the Swiss Federal Institute of Technology (Switzerland).

From 2003 to 2004 she was a La Jolla Interfaces in Science Training Program (LJIS) Postdoctoral Fellow at the University of California San Diego (UCSD) and The Scripps Research Institute (TSRI), where she worked on interdisciplinary projects in computer systems and computational chemistry. From 2005 to 2007, she was an Assistant Professor at the Computer Science Department of the University of Texas at El Paso (UTEP).

She joined the University of Delaware in 2007 as an Assistant Professor and was promoted to Associate Professor with tenure in 2012.

Taufer's research interests include scientific applications and their advanced programmability in heterogeneous computing (i.e., multi-core and many-core platforms, GPUs), performance analysis, modeling, and optimisation of multi-scale applications on heterogeneous computing, cloud computing, and volunteer computing, numerical reproducibility and stability of large-scale simulations on multi-core platforms, big data analytics and MapReduce.