Classification performance of the SVM algorithm combining nucleotide correlation of dinucleotides across five organisms. The figure plots sensitivity (true positive rate) as a function of 1-specificity (false positive rate). Values in paren- theses denote the area under the receiver operating characteristic curve ( auROC ) 

Context in source publication

Context 1

... analyses have uncovered that nucleosome positioning exhibits intrinsic DNA sequence preferences. Short-range dominance of nucleotide correlation in the nucleosome and linker DNAs was found by calculating the information redundancy. However, the information redundancy represents the accumulated nucleotide correlation corresponding to16 dinucleotides. To further clarify the nucleotide correlation affecting nucleosome positioning, the parameter F k was introduced to examine particular base correlation corresponding to every dinucleotide (see “ Materials and methods ” for details). The calculated results indicated that the distribution of particular base correlation is significantly different in the nucleosome and linker DNA regions through the S . cerevisiae , D . melanogaster , and C . elegans genomes (see ESM Figs. S7, S8, S9, S10, and S11). Besides, the profiles of base correlation of the 16 dinucleotides were slightly different between the normal nucleosome and the nucleosome consist- ing of a histone variant. More details can be founded in ESM. To efficiently confirm the relationship between base correlation and nucleosome positioning, the SVM classifier in terms of each base correlation corresponding to 16 dinucleotides was employed to discriminate the nucleosome and linker DNA sequences in eukaryotic genomes. To compare the performance of our model with other methods, the nucleosome positioning datasets of H . sapiens , O . latipes , C . elegans , C . albicans , and S . cerevisiae were retrieved from data published by Tanaka and Nakai (2009). For each organism, the dataset com- prised randomly extracted 1,000 nucleosomal and 1,000 linker DNA sequences. The parameter F k ( k = 0 ... 98), describing particular base correlation corresponding to the 16 dinucleotides, was calculated for each DNA sequence. Using a 1,584-element (99×16) vector as input vectors, the SVM classifier discriminating the nucleosome and linker DNA sequences was constructed. The performance of the classifier was measured using an independent validation procedure. In this procedure, the dataset is randomly divided into two subsets. One of them was trained by SVM and the other was tested. The performance of the SVM classifier was shown in Table 2. The model obtained a good performance with an average total accuracy of 76.0500 % in five organisms. To compare the performance of our model with other methods, the auROC was also calculated. Our SVM model obtained a mean auROC of 0.8353 (see Fig. 3). The result indicated that the prediction accuracy of our model is significantly higher than the method mentioned in the study of Tanaka and Nakai (2009) and is identical approximately to the study of Zhang et al. (2012). This finding suggested again that nucleotide correlation information is the important signature of nucleosome positioning. The information-theoretic method was applied to detect information on nucleotide correlation stored in the nucleosome and linker DNA sequence across the three organisms mentioned above. The results showed that information content of nucleotide correlation in the nucleosome and linker DNA regions is significantly different and that short-range nucleotide correlation in the nucleosome and linker DNA sequence is dominant. Short-range dominance of nucleotide correlation in the nucleosome and linker DNA regions is probably the major reason why many successful prediction models of nucleosome positioning in eukaryotic organisms were constructed in terms of sequence information of oligonucleotides or k -mer. Besides, the difference between normal nucleosome and H2A.Z-containing nucleosome was elucidated from the perspective of information theory. Next, the profiles of periodicity in the nucleosome and linker DNA for the S . cerevisiae , D . melanogaster , and C . elegans genomes indicated that the periodicities exist obviously and is species-specific. In D . melanogaster , the profile of the power spectrum was different between the H34-containing and the H2A.Z-containing nucleosome. Furthermore, the SVM model combining particular nucleotide correlation corresponding to the 16 dinucleotides was used successfully to discriminate the nucleosome and linker DNA sequences in H . sapiens , O . latipes , C . elegans , C . albicans , and S . cerevisiae . This application confirmed the importance of nucleotide correlation. Although the question of whether nucleosome positioning in vivo is determined by chromatin code is hotly debated, the fact that information stored in primary DNA sequence is an important determiner of nucleosome formation has been ...