Introduction

    Deep Belief Network is branch of Deep Learning stacking several layers of Restricted Boltzmann Machine. It produce good result on both supervised and unsupervised learning.This is a DBN classification tool in bioinformatic focus on predicting protein stability change caused by single-site amino acid mutation. Three methods can be used in this tool while you should provide at least sequence information and mutation information.

Deep Belief Network


    Single mutation information is required in this case. Such as position, original amino acid, mutated amino acid and original protein sequence.


A sequence with over 7 amino-acids is preferred.

    Mutated sequence will be compared to original sequence in this case. A set of mutations will be listed showing the change of protein stability.


A sequence with over 7 amino-acids is preferred.
Length of Sequences must be the equal. Miss values in both sequences could be labeled as " ~ "

    A text file of mutation information could be uploaded in this case. The mutation information should be input in specified format.






Detail

     As a method of Deep Learning, Deep Belief Network forward propagate first. Unlike a common Neural Network, DBN produces a better set of weights and biases by stacking several RBMs layer by layer instead of setting them randomly. This is called pre-training of a DBN. After that, DBN is fine-tuned by back-propagation which is the same as a regular Neural Network.

    Following gif shows how a regular DBN works.

Deep Belief Network





Department of Computer Science    
University of Missouri - Columbia    

Contact