A novel signal processing approach for the detection of copy number variations in the human genome

Abstract

caterina@mit.edu.

Estimation of correlations between copy-number variants in non-coding DNA

“…A common reference sequence was used to normalize each sequence in a particular batch. Array CGH data is typically noisy and contains genomic artifacts which were suppressed using a denoising procedure involving sequence decomposition into individual signal components and elimination of low-amplitude, high-frequency components, a process that also increased the data signal-to-noise ratio (SNR) [21]. Matched-filtering, a quasi-optimum pattern matching filtering method, was then applied to detect regions of dissimilarity between sequences and thus CNVs.…”

“…We have previously developed a methodology based on the matched-filter for detecting regions of pairwise similarity and dissimilarity between genomic sequences [20][21]. By definition, the matched-filter increases the signal-to-noise ratio (SNR) in regions of pairwise waveform similarity and decreases SNR in regions of mismatch.…”

“…As a waveform matching technique, matched-filtering treats discrete DNA sequences as continuous signals, potentially resulting in spurious spatial correlations between probes. However, we have previously shown that this approach does not introduce significant correlations in the filtered sequence [21]. In addition, the method strongly depends on the choice of the template sequence.…”

“…Finally, a CNV was called based on the frequency of its occurrence. The probability of a CNV at marker i was defined as the union of the probabilities of gain and loss at that marker [21]: $$Pr({\mathit{CNV}}_i)=\frac{{\sum }_j{\mathrm{italiclog}}_{2}false(ifalse)\ge {\mathrm{italiclog}}_{2}false(1\frac{3}{2}false)+{\sum }_m{\mathrm{italiclog}}_{2}false(ifalse)\le {\mathrm{italiclog}}_{2}false(1\frac{1}{2}false)}{n}$$ where j = 1, …, J is the number of sequences with gains above the threshold, m = 1, …, M the number of sequences with losses below the threshold, and n the total number of sequences in the sample. A 10% frequency was chosen as the threshold.…”

“…We present preliminary results on the estimation of correlated CNVs in a small set of non-coding, evolutionarily ultra-conserved genomic regions, detected in Array Comparative Genomic Hybridization (array CGH) data from healthy individuals. To detect these CNVs we applied a novel signal processing-based method that treats genomic sequences as continuous signals and uses a matched-filtering approach to identify regions of pairwise similarity and dissimilarity [21]. We show that there are predominantly local correlations between CNVs within the same chromosome, but that there are only limited correlations between CNVs in different chromosomes.…”