A Two-Stage Evolutionary Approach for Effective Classification of Hypersensitive Dna Sequences

Abstract

Hypersensitive (HS) sites in genomic sequences are reliable markers of DNA regulatory regions that control gene expression. Annotation of regulatory regions is important in understanding phenotypical differences among cells and diseases linked to pathologies in protein expression. Several computational techniques are devoted to mapping out regulatory regions in DNA by initially identifying HS sequences. Statistical learning techniques like Support Vector Machines (SVM), for instance, are employed to classify DNA sequences as HS or non-HS. This paper proposes a method to automate the basic steps in designing an SVM that improves the accuracy of such classification. The method proceeds in two stages and makes use of evolutionary algorithms. An evolutionary algorithm first designs optimal sequence motifs to associate explicit discriminating feature vectors with input DNA sequences. A second evolutionary algorithm then designs SVM kernel functions and parameters that optimally separate the HS and non-HS classes. Results show that this two-stage method significantly improves SVM classification accuracy. The method promises to be generally useful in automating the analysis of biological sequences, and we post its source code on our website.

Effective Automated Feature Construction and Selection for Classification of Biological Sequences

“…In particular, recent work has shown improved classification accuracies when using genetic algorithms (GAs), a class of EAs, to replace feature enumeration techniques in predicting promoter regions, HS sites, and splice sites in DNA, and even enzymatic activity in proteins [15], [16], [26]–[28], [38].…”

“…The splice site prediction problem is now considered a primary subtask in gene finding and is thus the subject of many machine learning methods [6] – [16] . Other prominent DNA analysis problems involve the identification of regulatory regions [17] , [18] through detection of binding sites of transcription factors [19] – [21] or detection of hypersensitive sites as reliable markers of regulatory regions [15] , [22] – [28] , identification of ALU sites [29] – [33] to understand human evolution and inherited disease [34] , [35] , and more.…”

“…In these kernels, limited-range dependencies between neighboring nucleotides are considered to encode features for the SVM. Concepts from evolutionary computation have been lately proposed to learn effective, possibly more complex, kernels for a particular sequence classification problem at hand [27] , [28] .…”

“…In many classification problems on biological sequences, research has shown that simple spectrum (compositional-based) features are not sufficient. Problems, such as predicting protein enzymatic activity, DNA hypersensitive sites, or RNA/DNA splice sites seem to necessitate complex local and distal features [11] , [13] , [14] , [25] – [28] , [38] . In particular, taking into account dependencies through features that encode correlations or simultaneous occurrences of particular -mers at different positions in a biological sequence is shown to be important for accurate detection of splice sites [14] – [16] .…”

“…Recognized early for their promise in addressing difficult optimization problems [77] , EAs have gained popularity for feature construction in different application settings [16] , [28] , [38] , [78] – [83] . In particular, recent work has shown improved classification accuracies when using genetic algorithms (GAs), a class of EAs, to replace feature enumeration techniques in predicting promoter regions, HS sites, and splice sites in DNA, and even enzymatic activity in proteins [15] , [16] , [26] – [28] , [38] .…”