Abstract by Bryce Hedelius
Physics and Astronomy
Dennis Della Corte
Protein Structure Prediction via Distance Mapping and Sequence Alignment
Although proteins are sequences of just 20 different amino acids, they are responsible for virtually every life sustaining-process. Proteins consistently fold the same way, but predicting the tertiary structure based on the sequence of amino acids remains one of the greatest unsolved problems in computational biochemistry. The state-of-the-art prediction method is a convolutional neural-network (CCN) produced by Google's Deepmind.
At BYU we are creating an improved prediction pipeline called ProSPr. An initial step is the setup of databases to train a CNN. Here we present our preliminary results and methodology in architecture and training of the CNN.