Analysis of Genomic and Proteomic Signals Using Signal Processing and Soft Computing Techniques

Abstract

Bioinformatics is a data rich field which provides unique opportunities to use computational techniques to understand and organize information associated with biomolecules such as DNA, RNA, and Proteins. It involves in-depth study in the
areas of genomics and proteomics and requires techniques from computer science,statistics and engineering to identify, model, extract features and to process data for analysis and interpretation of results in a biologically meaningful manner.In engineering methods the signal processing techniques such as transformation,filtering, pattern analysis and soft-computing techniques like multi layer perceptron(MLP) and radial basis function neural network (RBFNN) play vital role to effectively resolve many challenging issues associated with genomics and proteomics.
In this dissertation, a sincere attempt has been made to investigate on some challenging problems of bioinformatics by employing some efficient signal and soft computing methods. Some of the specific issues, which have been attempted are protein coding region identification in DNA sequence, hot spot identification in protein, prediction of protein structural class and classification of microarray gene expression data. The dissertation presents some novel methods to measure and to extract features from the genomic sequences using time-frequency analysis and machine intelligence techniques.The problems investigated and the contribution made in the thesis are presented here in a concise manner. The S-transform, a powerful time-frequency representation technique, possesses superior property over the wavelet transform and short time Fourier transform as the exponential function is fixed with respect to time axis while the localizing scalable Gaussian window dilates and translates. The S-transform uses an analysis window whose width is decreasing with frequency providing a frequency dependent resolution. The invertible property of S-transform makes it suitable for time-band filtering application. Gene prediction and protein coding region identification have been always a challenging task in computational biology,especially in eukaryote genomes due to its complex structure. This issue is resolved using a S-transform based time-band filtering approach by localizing the period-3 property present in the DNA sequence which forms the basis for the identification.Similarly, hot spot identification in protein is a burning issue in protein science due to its importance in binding and interaction between proteins. A novel S-transform based time-frequency filtering approach is proposed for efficient identification of the hot spots. Prediction of structural class of protein has been a challenging problem in bioinformatics.A novel feature representation scheme is proposed to efficiently
represent the protein, thereby improves the prediction accuracy. The high dimension and low sample size of microarray data lead to curse of dimensionality problem which affects the classification performance.In this dissertation an efficient hybrid feature extraction method is proposed to overcome the dimensionality issue and a RBFNN
is introduced to efficiently classify the microarray samples.