IMPROVING PROTEIN DISORDER PREDICTION USING ENSEMBLE OF EMBEDDING, CONVOLUTIONAL AND RECURRENT LAYERS

Abstract

Intrinsically Disordered Regions (IDRs) are the regions in proteins that do not posses well organized two dimensional or three dimensional structures at their physiological conditions. These regions exist extravagantly in each domain and concerned with wide range of protein functions. Perceiving this far reaching presence of these regions in proteins, prodded the improvement of computational strategies to discover more of them. Every current procedure can be arranged into techniques depending on evolutionary profiles produced from multiple sequence alignment and those depending on sequence information. The techniques dependent on evolutionary sequence profiles are progressively more precise than single sequence methods due to the fact that the evolutionary sequence profiles contain significant information relating to the absence or presence of preserved residues due to their respective functional and structural roles. However, the tedious count of sequence profiles restricts the wide pertinence of profile dependent methods. Hence, study was proposed to hypothesize a strategy to reduce the performance gap between sequence dependent methods and profile dependent methods. Here we showed a model with enhanced accuracy utilizing an ensemble of embedding, convolutional and bi-directional LSTM layers with for predicting disordered regions in proteins in contrast to already existing state of art techniques. Successful prediction will help research community for both the onset of diseases and restorative treatment in context to intrinsically disordered regions or proteins.