Slides -n._sitdykova_0
0 likes319 views
This article discusses a novel algorithm for filtering artificially duplicated reads from 454 pyrosequencing data. The algorithm operates directly on the flowgram data without converting to base calls. It uses a hierarchical clustering approach in "flow space" to group similar flowgrams. Benchmarking on a large 454 sequencing dataset showed the algorithm effectively removed duplicates while maintaining a high Jaccard index compared to clustering based on mapped reads. Removing duplicate reads improved metrics of genome assemblies such as N50.
Slides -n._sitdykova_0
- 1. Article “Filtering duplicate reads from 454 pyrosequencing data“ Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen Reporter: Nadiya Sitdykova 11.05.2013 Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 1 / 12
- 2. Motivation Artificially duplicated reads - duplicated reads, which don’t arise naturally, i.e. by chance trough sampling DNA molecules that start at identical positions or in rprtetive regions of a genome. lead to incorrect conclusions about coverage of read. Extremely sensetive to arificially duplicated reads applications of sequencing: de novo whole-genome sequencing metagenomics metatranscriptomics Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 2 / 12
- 3. Sources of artificial duplicates Emulsion PCR Droplets containing several empty beads and a single DNA molecule results in loading these beads with identical copies of the original DNA molecule. Background amplicon contamination Avoided by preventing cross-contamination of sequencing library samples. Signal cross-talk on the PTP sequencing device With the launch of the 454 Titanium chemistry, well cross-talk has been mimimized by metal coating of the PTP well surface. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 3 / 12
- 4. Flow space The native output format of 454 pyrosequencing is the binare standard flowgram format (*.sff). It contains the flowgram for each read, whereby each flowgram consists of a wequence of flow values representing base incorporations. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 4 / 12
- 5. Algorithm Novelty - operate only in flow space. Preclustering Create subsets of data: flowgrams from different subsets cannot be identified as duplicates of each other. Use a varying seed of at least 8 flows, starting with the first flow Take into account if the flow value was ’negative’(<0.5) or ’positive’(≥ 0.5) Require flowgrams in one precluster to start with the same homopolymer length For preclusters containing >2000 flowgrams gradually increase seed and split them up Hierarchical clustering Iterate through the files with the preclustered flowgrams and perform agglomerative clustering on one file at a time. Start: each cluster contain one flowgram. Calculate all pairwise distances between flowgrams. Two clusters with with smallest distance Determine consensus: calculate the per-flow median of flow values from all flowgrams in this cluster (quality-trimmed regions only). Update distance between new cluster and all other clusters Stop: smallest distance exceed a given stringency threshold Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 5 / 12
- 6. Distance between flowgrams Probability for a homopolymer length being equal to h when observing a flow value f : P(h|f ) = P(f |h) · P(h) P(f ) Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 6 / 12
- 7. Distance between flowgrams Assume that flowgrams fga and fgb are independent. Probability that the homopolymer lengths, hai and hbi , are equal, given two flow values, fai and fbi : P(hai = hbi |fai , fbi ) := 1, if fai or fbi > 5.5; 1, if fai and fbi > 2.5; 5 k=0 P(hai = k|fai ) · P(hbi = k|fbi ), else. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 7 / 12
- 8. Distance between flowgrams We take into accaount the 454 key and quality trimming information. So only infomative flow values are used. Assume that the flow values of one flowgram are not correlated. Define distance between two flowgrams: d(fga, fgb) := − log m i=l P(hai = hbi |fai , fbi ) m − (l − 1) = m i=l − log P(hai = hbi |fai , fbi ) m − (l − 1) l = max{left trimpoint(fga), left trimpoint(fgb)}, m = min{400, right trimpoint(fga), right trimpoint(fgb)} We take into account the 454 key and quality trimming information. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 8 / 12
- 9. Benchmarking Benchmarking dataset: 1270325 Dicentrarchuslabrax 454 GS FLX Titanium reads. Map it to corresponding reference scaffolds. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 9 / 12
- 10. Benchmarking The measure that compares two sets of cluster is Jaccard index: Jaccard := a a + b + c a - flowgram pairs that are correctly identified as duplicates b - flowgram pairs that are incorrectly not identified as duplicates c - flowgram pairs that are incorrectly identified as duplicates Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 10 / 12
- 11. Benchmarking Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 11 / 12
- 12. Benchmarking Additionally tested the effect of duplicate removal on assembly performance of the E.coli genome: Filtered reads assembled using Newbler. Score resulting assemblies using Mauve assembly metrics. Result: for both JATAC and cd-hit-454 the N50 increased from 106414bp to 126844bp. Susanne Balzer, Ketil Malde, Markus A. Grohme and Inge Jonassen (Reporter: Nadiya Sitdykova)Article “Filtering duplicate reads from 454 pyrosequencing data“ 11.05.2013 12 / 12