ABSTRACT
Existing DSP techniques are less efficient in terms of accuracy in exon prediction, computational complexity, and convergence performance. Complexity reduction and prediction accuracy, particularly in genomic signal-processing (GSP) applications, have continued to be an extreme research topic. In order to cope with both these problems, several less-complex adaptive techniques have been proposed for exon prediction. We have developed three categories of adaptive algorithms to accomplish this task. They are weight update variants of sign-based least mean square (LMS) algorithms, normalization-based adaptive algorithms, and logarithmic realization of normalized adaptive algorithms. As part of the first category, five algorithms, namely, LMS algorithm, least mean forth (LMF) algorithm, variable step size LMS (VSLMS) algorithm, least mean logarithmic squares (LMLS) algorithm, and least logarithmic absolute difference (LLAD) algorithm along with its signed variants are considered for development of various adaptive exon predictors (AEPs). In GSP, the capability of processing lengthy sequences in multiple iterations is presented by using adaptive techniques, and also their weight coefficients of input sequence can alter based on statistical nature.
LLAD outperforms the sign algorithm (SA) and also remains vigorous contrary to impulse interferences. Thus, this algorithm significantly improves the convergence performance, improves exon locating ability, and offers less computational complexity than LMS and all other adaptive variants presented in the first category. These techniques are further extended thru linking the sign algorithms like SA, sign regressor algorithm (SRA), and sign-sign algorithm (SSA). All these methods need merely 50% of multiplication computations as in their LMS corresponding versions, thus making them beneficial for practical applications.
In this chapter, we introduce AEPs designed to predict exon sites in DNA samples using multiple LMS-based adaptive filtering methods. These methods are further expanded to include signed realizations to decrease computational difficulty. This chapter also offers a performance analysis depending on various metrics such as sensitivity, specificity, precision, computational complexity issues, and convergence analysis, in relation to adaptive filtering methods. Also, a detailed discussion on various gene datasets considered for genomic sequence analysis is presented.
