Identify protein disorder from amino acid sequences with Machine learning

Abstract

Intrinsic protein disorder can predict a whole host of neurodegenerative diseases like Alzheimer's. Predicting protein disorder itself is best undertaken using computational methods. In this paper, a novel approach to predicting protein disorder using a Convolutional Neural Network (CNN) algorithm. The algorithm had found a 92% auc-roc score, which indicates the performance of a binary classification model. To extract features, the model had used OneHotEncoding, a technique that converts the sequential data into numerical values that is fed into the model. The data was also gathered from a variety of sources including the Protein Data Bank (PDB), The Disordered Protein Database (Disprot), and the Swiss protein database (Swissprot) [1]-[3]. This paper has a unique approach in the model built and distinguished itself from prior models based on the structure of the model and feature extraction that was performed. Using a tensorflow framework, the model used multiple convolutional layers of varying filter length and a final dense layer to enable the model to learn the features and predict associated outputs. Furthermore, the output from the initial stages was fed into a binary cross entropy classifier that gave the resulting judgement of order or disorder.