Myzus persicae Clone G006 Assembly

Transcription

1 Myzus persicae Clone G006 Assembly R. Chikhi, T. Derrien, F. Legeai October 8, Reads correction Although sequence qualities from Illumina technologies are known to be accurate, typical errors are substitutions and arise at a frequency %, especially at the 3 end of the reads [7]. Assembling reads that contains errors may lead to false positive overlaps of k-mers ( reads) or trigger false positive gaps in the final assembly. We therefore used a state-ofthe-art program, Quake [7], that is dedicated to identify and correct reads. Briefly, Quake uses a specific method to choose an appropriate coverage cutoff between trusted k-mers (those that are truly part of the genome) and erroneous k-mers based on weighting k-mer counts in the reads using the quality values assigned to each base. The main statistics of this correcting steps are summarized in table 1. 2 Assembly 2.1 Minia assembly The Minia assembler (v ) [4] was used to assemble the paired-end reads into contigs. It was executed in de Bruijn graph assembly mode, i.e. no scaffolds were created. The commandline parameters are k=71, this value was recommended following a run of kmergenie [3] and min abundance=43. The contigs were (s) scaffolded and (g) gap-filled by executing each operation two times as follows: s+g+s+g. The scaffolding software used is SuperScaffolder Number Validated Corrected Trimmed Total Library of fragmentmentsmentsments* fragments frag- frag- frag- cleaned Name MPA MPB S(PE) Table 1: Summary table of Quake corrections on Myzus libraries. (* each number correspond to the 2 mate files per library), reads number are in millions 1

2 [5], a modified version of SSPACE [2] that is still in development. SuperScaffolder version was executed with no other parameters than the read files and the insert size of each library. The gap-filling software used is GapCloser v1.12 from SOAPdenovo [11]. GapCloser was executed with default parameters. Only the two mate-pairs libraries were used in scaffolding and gap-filling. Minia and SuperScaffolder have been used on non-corrected reads. 2.2 Allpaths-LG2 assembly The AllPaths-LG (r40324) [13] assembler was runned using the default parameters as recommended by the authors. We provided the followings descriptor files : in groups.csv : group_name, library_name, file_name MPA, MPA, awilson_mpersicaeg006_ mpa_s_7_?.fastq MPB, MPB, awilson_mpersicaeg006_ mpb_s_8_?.fastq PE, PE, awilson_mpersicaeg006_ _s_6_?.fastq in libs.csv : library_name, project_name, organism_name, type, paired, frag_size, frag_stddev,insert_size, insert_stddev, read_orientation, genomic_start, genomic_end MPA, Myzus, Myzus persicae, jumping, 1,,, 5000,500, outward,, MPB, Myzus, Myzus persicae, jumping, 1,,, 2000,200, outward,, PE, Myzus, Myzus persicae, fragment, 1, 200, 20,,, inward,, 2.3 Abyss assembly The Abyss assembler v1.3.2 [15] has been used on the corrected reads using different kmer size (k=41, 64 and 91). 3 Metrics We computed standard statistics on the AllPaths-LG2, Minia and Abyss assemblies, scaffolds (table 2) and contigs (table 3). These metrics have been calculated using the script assemblathon stats2.pl, used during the Assemblathon contest [6]. The expected size of the genome has been fixed at 350MBP, the number of consecutive N used to split scaffolds in contigs is 3. All these metrics have been calculated on scaffolds larger than 1000 bp. 3.1 Conclusion Minia gives better metrics, especially a better scaffold NG50, and less N in scaffolds. This better performance is not essentially due to a Gap closing step or a better scaffolding because the contigs show also a better NG50. In comparison with the other assemblers, the metrics given by Abyss are not good (number of scaffolds, NG50,... ). Thus, we decided to discard Abyss from further analyses and compares the AllPaths-LG and Minia assemblies, only. 2

3 Metric Allpaths Minia Abyss (k=41) Abyss (k=64) Abyss (k=91) Number of scaffolds Total size of scaffolds Total scaffold length as percentage of known 99.5% 99.3% 96.1% 103.8% 99.7% genome size Longest scaffold Shortest scaffold Number of scaffolds > 10K nt 1858 (43.2%) (37.9%) (39.6%) 5128 (41.0%) 8916 (32.1%) Number of scaffolds > 100K nt 788 (18.3%) 673 (18.3%) 976 (8.1%) 1031 (8.2%) 246 (0.9%) Number of scaffolds > 1M nt 38 (0.9%) 67 (1.8%) 0 (0.0%) 0 (0.0%) 0 (0.0%) Mean scaffold size Median scaffold size N50 scaffold length L50 scaffold count NG50 scaffold length LG50 scaffold count N50 scaffold - NG50 scaffold length difference scaffold %A scaffold %C scaffold %G scaffold %T scaffold %N scaffold N nt Table 2: Scaffolds metrics Metric Allpaths Minia Abyss (k=41) Abyss (k=64) Abyss (k=91) Percentage of assembly in scaffolded contigs 95.4% 96.2% 93.5% 88.8% 90.8% Percentage of assembly in unscaffolded contigs 4.6% 3.8% 6.5% 11.2% 9.2% Average number of contigs per scaffold Average length of breaks (3 or more Ns) between contigs Number of contigs Number of contigs in scaffolds Number of contigs not in scaffolds Total size of contigs Longest contig Shortest contig Number of contigs > 500 nt (98.8%) (92.4%) (87.0%) (94.1%) (91.8%) Number of contigs > 1K nt (96.1%) 8655 (80.5%) (79.2%) (86.7%) (79.3%) Number of contigs > 10K nt 5915 (46.2%) 3983 (37.0%) (22.8%) (31.7%) 4173 (3.9%) Number of contigs > 100K nt 870 (6.8%) 1072 (10.0%) 9 (0.0%) 65 (0.2%) 1 (0.0%) Number of contigs > 1M nt 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) Mean contig size Median contig size N50 contig length L50 contig count NG50 contig length LG50 contig count N50 contig - NG50 contig length difference contig %A contig %C contig %G contig %T contig %N contig N nt Table 3: Contigs metrics 3

4 Metric Allpaths Minia number of reads number of mapped reads (86.67%) (87.72%) number of aligned reads >1 times (7.80%) (15.01%) number of properly paired reads (83.11%) (79.79%) number of reads with itself and mate mapped () (85.57%) singletons (1.89%) (2.15%) with mate mapped to a different chr (1.35%) (2.09%) number of improperly paired reads on the same chr (0.03%) (3.70%) Table 4: PE mapping statistics Metric Allpaths Minia number of reads number of mapped reads (63.96%) (64.49%) number of aligned reads >1 times (1.42%) (1.67%) number of properly paired reads (47.19%) (45.69%) number of reads with itself and mate mapped (57.00%) (57.06%) singletons (6.95%) (7.44%) with mate mapped to a different chr (6.17%) (6.94%) number of improperly paired reads on the same chr (3.64%) (4.43%) 4 Remapping Reads 4.1 Protocol Table 5: MPA mapping statistics The corrected reads have been remapped on the genome using bowtie2 [9]. We used the option -X 500 for the pair ends mapping and the options rf for the mate pairs reads, we put the option -X to 8000 and 5000 for the MPA and MPB libraries, respectively. For the mapping statistics, we used the option flagstat of samtools [10]. For the coverage analysis, we used the tool genomecov of bedtools [14] and R scripts. 4.2 Mapping statistics An overview of the tables 4, 5, 6 below shows that both assemblers give very similar results, but it appears that Allpaths-LG shows better statistics on properly paired mapping for each library. It appears also that there is 3 times more reads with multiple putative locations found in the Minia assembly, although there is just few more mapping reads. 4

5 Metric Allpaths Minia number of reads number of mapped reads (75.23%) (75.79%) number of aligned reads >1 times (2.51%) (2.31%) number of properly paired reads (63.72%) (61.66%) number of reads with itself and mate mapped (71.88%) (71.84%) singletons (3.35%) (3.95%) with mate mapped to a different chr (6.84%) (7.55%) number of improperly paired reads on the same chr (1.32%) (2.63%) Table 6: MPB mapping statistics Min. 1st Qu. Median Mean 3rd Qu. Max Table 7: Allpaths depth coverage by scaffolds 4.3 Analysis of the coverage Allpaths Following the mapping reads statistics we are expecting a depth of coverage of 140x. But the median coverage by scaffold given by genomecov is slightly lower as described (cf table 7). This might be explained by a contamination of bacterial population or recent transposable elements regions in the nuclear genome. Thus, we plotted the GC percent and coverage by scaffolds (see figure 1), and observe a distinct cloud (in the orange circle) with a coverage higher than 1000x, suspected as being putatively bacterial contamination, and discard these scaffolds for further analyses. This set includes 222 scaffolds covering bp. Among these 222 scaffolds, 173 return at least one hit when compared to Genbank Bacterial division, 171 matchs especially buchnera aphidicola. (see table 8) Minia As previously, we obtained a median coverage lower as described in table 9. The figure 2, we can also point out a distinct cloud (in orange) with a coverage higher than 1000x, we also discard them from the set for further analyses. This set includes 35 scaffolds covering bp. 23 of these 35 scaffolds return at least one hit when compared to Genbank Bacterial division, 22 matchs especially Buchnera aphidicola. (see table 10) 5

6 Figure 1: Plot of the coverage x GC percent of the allpaths scaffolds, the orange circle highlights the scaffolds with high coverage Bacterial hit # scaffolds Buchnera aphidicola strȧk (Acyrthosiphon kondoi) 112 Buchnera aphidicola str. JF98 (Acyrthosiphon pisum) 24 Buchnera aphidicola str. Ua (Uroleucon ambrosiae) 11 Buchnera aphidicola str. JF99 (Acyrthosiphon pisum) 7 Buchnera aphidicola str. TLW03 (Acyrthosiphon pisum) 4 Buchnera aphidicola str. 5A (Acyrthosiphon pisum) 3 Buchnera aphidicola str. Sg (Schizaphis graminum) 2 Buchnera aphidicola str. LL01 (Acyrthosiphon pisum) 2 Buchnera aphidicola str. APS (Acyrthosiphon pisum) 1 Buchnera aphidicola (Myzus persicae) hupa-rpoc intergenic spacer 1 Buchnera aphidicola DNA polymerase III beta subunit (dnan) gene 1 Buchnera aphidicola (Diuraphis noxia) plasmid pleu-dn(usa8) 1 Buchnera aphidicola (Acyrthosiphon kondoi) 1-deoxy-D-xylulose 5-phosphate reductoisomerase (dxr) gene, 1 Bacterium IS422 gene for 16S rrna 1 Bacillus subtilis BSn5 1 Azotobacter vinelandii isolate DNA S ribosomal RNA gene 1 No hit 49 Table 8: Hits of the highly covered AllPaths scaffolds against the Genbank Bacterial division Min. 1st Qu. Median Mean 3rd Qu. Max Table 9: Minia depth coverage by scaffolds 6

7 Figure 2: Plot of the coverage x GC percent of the Minia scaffolds longer than 1000bp, the orange circle highlights the scaffolds with high coverage Bacterial hit # scaffolds Buchnera aphidicola str. Ak (Acyrthosiphon kondoi) 16 Buchnera aphidicola str. Ua (Uroleucon ambrosiae) 2 Buchnera aphidicola str. JF98 (Acyrthosiphon pisum) 2 Buchnera aphidicola str. JF99 (Acyrthosiphon pisum) 1 Buchnera aphidicola anthanilate syntyhase component I (trpe) and anthranilate synthase component II (trpg) genes 1 Bacillus subtilis BSn5 1 No hit 12 Table 10: Hits of the highly covered Minia scaffolds against the Genbank Bacterial division 7

8 evalue no hit unique hit multiple hits Median Mean Max. 1e e e e Table 11: Blast hits at different e-value threshold while comparing the 3369 BUSCO drosophila melanogaster set to the Minia assembly evalue no hit unique hit multiple hits Median Mean Max. 1e e e e Table 12: Blast hits at different e-value threshold while comparing the 3369 BUSCO drosophila melanogaster set to the Allpaths assembly 5 Mapping BUSCO proteins 5.1 Protocol We extracted the proteins from drosophila melanogaster from the OrthoDB BUSCO Arthopoda set (ftp://cegg.unige.ch/orthodb6/busco/). Firstly, we aligned this protein set by blast on the genome and retrieve the hits with an e-value lesser than 1e Blast hits We used different threshold for filtering blast results, and observe at each threshold, the number of missing proteins, or proteins with unique or multiple matchs. The results are reported in the tables 11 and Completion of the proteins For each of the 2440 and 2432 proteins having a hit with respectively Allpaths and Minia assemblies, we aligned the BUSCO protein to the corresponding scaffold using GeneWise [1]. For analyzing the completion of the predicted peptides in the Myzus genome and the BUSCO set, we simply compared the size of the proteins, and plotted the empirical cumulative distribution frequencies (ecdf), as presented in figures 3 and 4. We observe that the plots are very similar. For both assembly, 50% of the proteins are at least complete at 78% (red lines), and 30% of the proteins are 90% complete (orange lines). 8

9 Figure 3: Cumulative distribution of the BUSCO drosophila melanogaster set completion on the Allpaths-LG genome 5.4 Conclusion Interestingly only almost 2 thirds of the BUSCO proteins have been successfully anchored on the Myzus persicae genome. None assembly give better results than the other. 6 Mapping ACYPI proteins 6.1 Protocol We used exactly the same protocol using the pea aphid proteins from AphidBase. 6.2 Blast hits We used different threshold for filtering blast results, and observe at each threshold, the number of missing proteins, or proteins with unique or multiple matchs. The results are reported in the tables 13 and 14. 9

10 Figure 4: Cumulative distribution of the BUSCO drosophila melanogaster set completion on the Minia genome evalue no hit unique hit multiple hits Median Mean 3rd qu. Max. 1e e e e Table 13: Blast hits at different e-value threshold while comparing the pea aphid proteins set to the Allpaths-LG assembly evalue no hit unique hit multiple hits Median Mean 3rd qu. Max. 1e e e e Table 14: Blast hits at different e-value threshold while comparing the pea aphid proteins to the Minia assembly 10

11 Figure 5: Cumulative distribution of the pea aphid protein set completion on the Allpaths-LG genome 6.3 Completion of the proteins The and proteins having a hit with respectively Allpaths and Minia assemblies, we aligned the pea aphid proteins to the corresponding scaffold using GeneWise [1]. As previously, we plotted the cumulative distribution of the ratio protein size in figures 5 and 6. plots are very similar. For both assembly, 50% of the proteins are at least complete at 96.5% (red lines), and 64% of the proteins are 90% complete (orange lines). 6.4 Conclusion On both genomes, lot of pea aphid proteins were unmapped on the Myzus genome, especially using a high threshold. Also, lot of proteins were not uniquely mapped. For most of the proteins, anchoring their best hits gives a close to complete prediction. 11

12 Figure 6: Cumulative distribution of the pea aphid set completion on the Minia genome Min. 1st Qu. Median Mean 3rd Qu. Max. Sum Table 15: Statistics on the Allpaths-LG scaffolds including buchnera genome parts 7 Buchnera genome 7.1 Analysis of the buchnera sequences in the genomes We compared the genome sequences to the Buchnera aphidicola str. LSR1 (Acyrthosiphon pisum), whole genome shotgun sequence (Accession number : NZ ACFK ) using blastn (e-value threshold : 1e-20). In the Allpaths- LG genome, we found 168 scaffolds that are including a buchnera genome part, all of them are very short (see table 15), but the sum of all these scaffolds is lower than the expected buchnera size ( bp for the pea aphid LSRA strain). Contrarily, when analyzing the 34 minia scaffolds that include part of buchnera we found that their size distribution (see table 16) is much larger. Surprisingly, some scaffolds larger than the buchnera genome include the buchnera genome (see table 16). 12

13 Min. 1st Qu. Median Mean 3rd Qu. Max. Sum Table 16: Statistics on the Allpaths-LG scaffolds includin buchnera genome parts Figure 7: Dot plot of the buchnera scaffolds 1 aligned to the buchnera pea aphid strain LSR1 7.2 New buchnera genome assembly As a result, none of the assemblies were able to predict the complete buchnera genome. The reason is that this genome have a very high coverage compared to the nuclear genome, and because there is a lot of reads, there is also lot of errors creating very complex regions in the De Bruijn graph and prevent the assemblers to achieve their traversal. Thus, we proceed to a specific assembly of the buchnera genome using Minia and increasing the min abundance to 20 and kmer size threshold to 91 in order to remove a large part of the errors. This stringency is too high to build the nuclear genome because its coverage is too low, but is is efficient to assemble the buchnera genome. Indeed using blast we retreve the scaffolds that were similar to the buchnera genome, and found only 2 scaffolds with respective size of and These 2 scaffolds correspond to 2 distinct parts of the buchnera genome and cover them almost completely (see figures 7 and 8). 8 Duplications 8.1 Protocol We used Nucmer algorithm from the Mummer 3.22 package [8] to compare all the scaffolds against each other in order to find similar regions which might be due to assembly artefacts. We were considering as putatively artefactually duplicated regions (PADR), if they were larger than 1000bp with a percentage 13

14 Figure 8: Dot plot of the buchnera scaffolds 1 aligned to the buchnera pea aphid strain LSR1 of similarity higher than 90. On the Minia genome we found 4415 PADR, corresponding to bp, and 4662 PADR corresponding to bp in the Allpaths-LG genome. Thus, we removed scaffolds covered by 70% by a PADR. As a result we discarded 184 scaffolds from the AllPaths-LG assembly and 115 from the Minia assembly. 9 Gap Closing and final assembly 9.1 Protocol As a conclusion, we propose to use the AllPaths assembly without the scaffolds with a coverage higher than 1000x, without scaffolds similar to the buchnera aphidicola genome and without scaffolds covered by a putative duplicated regions. We perform the gap closing (i.e. covering the N between the contigs in scaffolds using the reads), using Gap Closer v1.12 from soapdenovo2 [12] using the 3 libraries. 9.2 Results As a result we obtained clone G006 assembly with the metrics summarized in table Conclusion We used various metrics to select the best assembly among the 5 assemblies computed using the Myzus G006 clone reads. The Abyss assemblies have bad general metrics, and we compared AllPaths-LG and Minia using other different metrics. Although they gave very similar results, in particular about the protein mapping statistics, it appears that the AllPaths assembly has a lower level of 14

15 Metric G006 assembly Number of scaffolds 4022 Total size of scaffolds Longest scaffold Shortest scaffold 959 Number of scaffolds > 500 nt % Number of scaffolds > 1K nt % Number of scaffolds > 10K nt % Number of scaffolds > 100K nt % Number of scaffolds > 1M nt % Mean scaffold size Median scaffold size 7170 N50 scaffold length L50 scaffold count 224 scaffold %A scaffold %C scaffold %G scaffold %T scaffold %N 0.53 scaffold N nt Table 17: Scaffolds metrics redundancy (reflected by the number of PE remapping statistics) but a larger region of putative artefactually duplicated regions. Moreover, Minia shows up some large nuclear scaffolds including parts of buchnera genome, while Allpaths- LG created separated scaffolds for buchenra genome. Both assemblies were not able to assemble completely the buchnera genome, and we did a new specific assembly to achieve this goal. As a result, we retrieve almost the complete buchnera sequence in 2 large scaffolds. References [1] E. Birney, M. Clamp, and R. Durbin. GeneWise and Genomewise. Genome Res., 14(5): , May [2] M. Boetzer, C. V. Henkel, H. J. Jansen, D. Butler, and W. Pirovano. Scaffolding pre-assembled contigs using SSPACE. Bioinformatics, 27(4): , Feb [3] R. Chikhi and P. Medvedev. Informed and automated k-mer size selection for genome assembly. Bioinformatics, Jun [4] Rayan Chikhi and Dominique Lavenier. Localized genome assembly from reads to scaffolds: practical traversal of the paired string graph. In Springer, editor, WABI 2011, Sarrebruck, Germany, August [5] Rayan Chikhi and Delphine Naquin. Graph-based scaffolding for nextgeneration sequencing. In JOBIM, [6] D. Earl, K. Bradnam, J. St John, A. Darling, D. Lin, J. Fass, H. O. Yu, V. Buffalo, D. R. Zerbino, M. Diekhans, N. Nguyen, P. N. Ariyaratne, W. K. Sung, Z. Ning, M. Haimel, J. T. Simpson, N. A. Fonseca,?. Birol, T. R. Docking, I. Y. Ho, D. S. Rokhsar, R. Chikhi, D. Lavenier, G. Chapuis, D. Naquin, N. Maillet, M. C. Schatz, D. R. Kelley, A. M. Phillippy, S. Koren, S. P. Yang, W. Wu, W. C. Chou, A. Srivastava, T. I. Shaw, J. G. 15

16 Ruby, P. Skewes-Cox, M. Betegon, M. T. Dimon, V. Solovyev, I. Seledtsov, P. Kosarev, D. Vorobyev, R. Ramirez-Gonzalez, R. Leggett, D. MacLean, F. Xia, R. Luo, Z. Li, Y. Xie, B. Liu, S. Gnerre, I. MacCallum, D. Przybylski, F. J. Ribeiro, S. Yin, T. Sharpe, G. Hall, P. J. Kersey, R. Durbin, S. D. Jackman, J. A. Chapman, X. Huang, J. L. DeRisi, M. Caccamo, Y. Li, D. B. Jaffe, R. E. Green, D. Haussler, I. Korf, and B. Paten. Assemblathon 1: a competitive assessment of de novo short read assembly methods. Genome Res., 21(12): , Dec [7] D. R. Kelley, M. C. Schatz, and S. L. Salzberg. Quake: quality-aware detection and correction of sequencing errors. Genome Biol., 11(11):R116, [8] S. Kurtz, A. Phillippy, A. L. Delcher, M. Smoot, M. Shumway, C. Antonescu, and S. L. Salzberg. Versatile and open software for comparing large genomes. Genome Biol., 5(2):R12, [9] B. Langmead and S. L. Salzberg. Fast gapped-read alignment with Bowtie 2. Nat. Methods, 9(4): , Apr [10] H. Li, B. Handsaker, A. Wysoker, T. Fennell, J. Ruan, N. Homer, G. Marth, G. Abecasis, and R. Durbin. The Sequence Alignment/Map format and SAMtools. Bioinformatics, 25(16): , Aug [11] R. Li, H. Zhu, J. Ruan, W. Qian, X. Fang, Z. Shi, Y. Li, S. Li, G. Shan, K. Kristiansen, S. Li, H. Yang, J. Wang, and J. Wang. De novo assembly of human genomes with massively parallel short read sequencing. Genome Res., 20(2): , Feb [12] R. Luo, B. Liu, Y. Xie, Z. Li, W. Huang, J. Yuan, G. He, Y. Chen, Q. Pan, Y. Liu, J. Tang, G. Wu, H. Zhang, Y. Shi, Y. Liu, C. Yu, B. Wang, Y. Lu, C. Han, D. W. Cheung, S. M. Yiu, S. Peng, Z. Xiaoqian, G. Liu, X. Liao, Y. Li, H. Yang, J. Wang, T. W. Lam, and J. Wang. SOAPdenovo2: an empirically improved memory-efficient short-read de novo assembler. Gigascience, 1(1):18, [13] I. Maccallum, D. Przybylski, S. Gnerre, J. Burton, I. Shlyakhter, A. Gnirke, J. Malek, K. McKernan, S. Ranade, T. P. Shea, L. Williams, S. Young, C. Nusbaum, and D. B. Jaffe. ALLPATHS 2: small genomes assembled accurately and with high continuity from short paired reads. Genome Biol., 10(10):R103, [14] A. R. Quinlan and I. M. Hall. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics, 26(6): , Mar [15] J. T. Simpson, K. Wong, S. D. Jackman, J. E. Schein, S. J. Jones, and I. Birol. ABySS: a parallel assembler for short read sequence data. Genome Res., 19(6): , Jun