Deep learning allows computational models with several processing layers to learn multiple degrees of abstraction for data representations. These techniques have vastly enhanced the state-of-the-art in speech recognition, visual object recognition, object detection, and a variety of other fields like drug development and genomics. Deep learning uses the backpropagation algorithm to show how a machine should adjust its internal parameters that are used to compute the representation in each layer from the representation in the previous layer, revealing intricate structure in massive data sets. When it comes to object recognition, the deep learning algorithm can potentially detect what the human eye cannot, and faster. The deep learning application to biology can help to make predictions with conformational analysis, the study of the different energy levels associated with the various behaviors of a molecule. The idea here is to then apply image processing principles to the data from proteins, rather than use a traditional photograph. Currently, analysis on conformations determine the current state of behaviors at nanosecond or microsecond data. Deep learning can aid in recognizing patterns of behaviors over a large period of time, automating this process for faster discovery and to uncover patterns unrecognized before. We present an exploratory project to address an approach to applying deep learning for object detection algorithms in order to investigate behaviors of proteins, in a way that has not been practiced before. We have chosen to work with 2DRMSD plots which are currently being used to capture conformational substates. Using the TensorFlow open-source library, we show a technique object detection applied to 2DRMSD plots of MutS, a large protein involved in DNA repair. This thesis investigates the extent to which the machine learning algorithms for image processing can be applied to analyze graphical data.
"},{"label":"Degree","value":["Bachelor of Arts with Departmental Honors"]},{"label":"Author-supplied Keywords","value":["Deep learning","Conformational analysis","Object detection"]},{"label":"Administrative Note","value":"Drupal Node ID: 24254