Digital Signal Processing in Protein Secondary Structure Prediction Debasis Mitra

Considerable research effort has been devoted to predicting the secondary structure of proteins from their amino acid sequences. Despite the plethora of prediction techniques, present methods typically have 76% approximate level of accuracy on an average. Thus, there is a considerable room for improvement. We present here a novel automated approach for the secondary structure prediction based on the Digital Signal Processing (DSP) techniques. DSP is an engineering discipline concerning the creation, manipulation and analysis of digital signals. Our technique involves two DSP operators, Convolution and Deconvolution, for the purpose of predicting secondary structures. We use some mappings between an amino acid sequences and the corresponding numerical time-series or “signals” that are processed. Convolution is a method of applying a filter on an incoming signal, producing an outgoing signal. Deconvolution is the inverse operation of convolution and permits the filter to be recovered if the outgoing signal and the incoming signal are known. Our method predicts three states (helix, strand, and coil) for the secondary structure. We presume that each protein has a corresponding filter, which when convolved with the incoming signal (mapped from the primary structure) produces the outgoing signal (mapped from the secondary structure). It is our contention that the unknown secondary structure of a target protein can be predicted by using the appropriate digital filter for a base protein of a significant amino acid sequence-similarity and whose secondary structure is known. Our work presented here attempts to corroborate that contention and experiments with some methodologies towards that direction.