Active Research Projects
Dr. Yaohang Li
Department of Computer Science
North Carolina A&T State University
High Resolution ab initio Protein Folding
The successful ab initio
protein structure prediction depends on the surmounting of three
efforts: (1) formulating an accurate and sensitive scoring function that
can lead the search process to the global minimum in the protein folding
energy landscape; (2) devising efficient moves (conformation changes)
toward the native conformation; and (3) developing a global optimization
algorithm that can efficiently escape from the deep local minima and
converge to the global energy minimum. Among these three efforts,
building an accurate and sensitive scoring function is of the most
importance. However, just like many other computational biology
problems, developing a sensitive and accurate scoring function is a very
difficult and even formidable job. In reality, even though many scoring
functions based on various criteria, such as energy, statistics,
secondary structure, loops, or contact pairs, have been proposed,
currently there does not exist a reliable and general scoring function
that can always drive a search to a native fold, and there is no
reliable and general global optimization method under these scoring
functions that can sample the conformation space adequately to guarantee
a significant fraction o near-natives (<3.0 A RMSD from the experimental
structure).
We seek to develop novel Monte Carlo approach to address the
issue of the insensitivity in the current existing protein folding
scoring functions and perform convergence analysis of the Monte Carlo
sampling process. We are expected to develop protein modeling software
tools to improve the resolution of ab initio protein structure
prediction. I am collaborating with Dr. Andrey Gorin at
Oak Ridge National Laboratory and Dr.
Charlie Strauss at Los Alamos National
Laboratory.
Markov Chain Monte Carlo
Collaborating with Drs. Andrey Gorin and Vladimir Protopopescu at
Oak Ridge National Laboratory, we have
developed two new stochastic global optimization/sampling methods.
The first one is called Accelerated Simulated Tempering (AST), which
intends accelerate the simulated tempering scheme with random walks
executed on a temperature ladder with various transition step sizes. By
suitably choosing the length of the transition steps, the accelerated
scheme enables the search process to execute large jumps and escape
entrapment in local minima, while retaining the capability to explore
local details, whenever warranted. Our simulations confirm the expected
improvements and show that the accelerated simulated tempering scheme
has a much faster convergence to the target distribution than Geyer and
Thompson’s simulated tempering algorithm and exhibits accuracy
comparable to the simulated annealing method.
The second
method is a population-based Markov Chain Monte Carlo approach,
so-called hybrid PT/SA scheme, which combines Parallel Tempering (PT)
and Simulated Annealing (SA) methods and is suitable for large-scale
parallel computing systems. Within the hybrid PT/SA scheme, a composite
system with multiple conformations is evolving in parallel in a
temperature ladder with various transition step sizes. The SA process
uses a cooling scheme to lead the temperature values in temperature
ladder down to the target temperature. The PT scheme is employed to
reduce the equilibration relaxation time of the composite system at a
particular temperature ladder configuration in the SA process. The
hybrid PT/SA method can reduce the waiting time in deep local minima and
thus leads to a more efficient sampling capability on high-dimensional
complicated objective function landscapes. Compared to the approaches PT
and parallel SA with the same temperature ladder, transition step sizes,
and cooling scheme (parallel SA) configurations, our preliminary results
obtained with the hybrid PT/SA method confirm the expected performance
improvements.
Sensor Grid
Many
different types of sensors are employed to monitor the different
environmental attributes contributing to the climate change. For
example, seismic sensors are deployed to monitor the seismic activities
under the ocean and temperature sensors and sea level sensors monitor
the changes in temperature and sea level in the ocean. While individual
sensors provide some insights about the ongoing events, it is very
important to consider the signals from different sensors collectively
for detecting climate change events. Patterns detected from the
individual sensors may look just normal in isolation. However, the
temporal relations among them across geospatially distributed sensors
may better indicate an important class of global events that may have
not been apparent from the individual stream analysis. Mining individual
stream data has been a subject of a large number of studies. However,
studies on mining spatio-temporal patterns across multivariate stream
data are very limited. In this project, we intend to address the
following three issues:
1. How the heterogeneous, geographically distributed sensors within a
sensor grid can be located, accessed, filtered, and integrated for a
particular study?
2. How the large amount of data can be collected and analyzed?
3. How a multitude of analysis components, such as statistical,
clustering, visualization, and classification tools can be applied to
mine the sensor data?
This project is supported by National
Oceanic & Atmospheric Administration (NOAA) through the
ISET center at North Carolina
A&T State University. I am collaborating with Mark Govett at NOAA,
Jaewoo Kang and Vincent Freeh at North Carolina State University, and
Albert Esterline at NCAT on this project.
Grid-based Monte Carlo and Quasi-Monte Carlo
Monte Carlo applications are widely perceived as computationally intensive but naturally parallel. Therefore, they can be effectively executed on the grid using the dynamic bag-of-work model. We improve the efficiency of the subtask-scheduling scheme by using an N-out-of-M strategy, and develop a Monte Carlo-specific lightweight checkpoint technique, which leads to a performance improvement for Monte Carlo grid computing. Also, we enhance the trustworthiness of Monte Carlo grid-computing applications by utilizing the statistical nature of Monte Carlo and by cryptographically validating intermediate results utilizing the random number generator already in use in the Monte Carlo application. All these techniques lead to a high-performance grid-computing infrastructure that is capable of providing trustworthy Monte Carlo computation services. These techniques can be also extended to quasi-Monte Carlo applications. Dr. Michael Mascagni is my key collaborator in this project.
Biologically-inspired Methods
Biological systems are remarkably adaptive and robust in complex real-world environments. For this reason, in this project, we are exploring biologically-inspired approaches to system adaptation, fault-tolerance and reconfiguration. Our approach is theoretical and application driven. On the theoretical side, we will explore the phenomena of self-adaptation and reconfiguration/organization in natural biological systems, and then develop a theoretical framework for self-reconfigurable systems. On the application side, we will design and evaluate biomimetic mechanisms and algorithms for future metamorphic autonomous systems. This project is funded by National Science Foundation, RISE program. More information can be found at the cooperative research center website.