This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. Primary support for the subproject and the subproject's principal investigator may have been provided by other sources, including other NIH sources. The Total Cost listed for the subproject likely represents the estimated amount of Center infrastructure utilized by the subproject, not direct funding provided by the NCRR grant to the subproject or subproject staff. Biomedical Research has been revolutionized with technological advances leading to massive accumulation of data. All this data now needs to be mined in order to draw actionable insights into the various biological processes. Complex machine learning algorithms are being developed to perform automated analyses of these large datasets and to come up with robust models that explain the observed data. Such models are then used to identify patterns in data that enable solving of challenging decision problems like diagnosis and prognosis of disease. Our research involves one such class of algorithms called Hidden Markov Models, which are used extensively in sequential data mining problems in Biology. Our particular focus is on development of novel algorithms for identification and quantification of protein sequences in complex biological samples using data that comes out of mass spectrometers. Such analysis will lead to molecular characterization of target conditions like diseased states. Our algorithms involve learning models from large training datasets and are computationally intensive. Additionally, in order to learn a robust model that will perform well across a variety of future test data, we are proposing to perform large-scale experiments with different model topologies and features, and require learning of hundreds of different models worth many days of number-crunching work. However, the entire experimentation can be parallelized trivially since all the models can be learned independently from each other and hence, the need for computing machines that can run multiple jobs in parallel. Our algorithms (homegrown) have been implemented using Python programming language and can take advantage of presence of multiple processing units or cores. After speaking with consultants at PSC, we were suggested that the Blacklight machines are most suitable for our needs.
Showing the most recent 10 out of 292 publications
