skoren
commented
4 years ago -
Both those assemblies and parameters seem reasonable, I'll just note that 175mb assemblies of ISO1 aren't that much larger than CLR. We typically get a 160mb assembly for the old ISO1 data (e.g.: https://github.com/marbl/canu-regression/blob/master/results/2018-10-19-1728/drosophila_melanogaster.pacbio.p5c3.report).
-
That doesn't make much sense, it seems almost like you're assembling both haplotypes together. That would explain both the assembly size and the continuity drop. I do see lots of reads with low coverage not present in the other reports which again is consistent with both haplotypes being mixed together (at low coverage). However, the histogram and the coverage don't support this, I'd expect to see a second peak in the histogram. Did you clean up everything from the old run when you re-tried the assembly? Can you share the ISO1 binned reads only to run locally.
-
The hifi option is still experimental and has some limitations in 1.9 (for example the gfa is not properly updated for the uncompressed space). However, it should produce better assemblies in terms of repeat resolution than pacbio-corrected, at least on complex genomes. I'm not sure there is enough complexity in drosophila for it to matter. The error rate for the hifi is not actually 2.5%, we just compute overlap out to that limit in case they can be corrected. By default it will only use overlaps below 0.03% error, much lower than used with pacbio-corrected. The 24kb library is much noisier than the 19kb library, in my experience, so you may be able to run the 19kb library with an even lower error rate (0%) by adding
batOptions="-eg 0.0 eM 0.0 -dg 3 -db 3 -dr 1 -ca 50 -cp 5", that might also get rid of the mis-assembly (but I expect it won't work on the 24kb library).
harsh-shukla
Hi Sergey,
I am trying to assemble these different HiFi libraries (https://www.ncbi.nlm.nih.gov/sra/?term=pacbio+HiFi+A4+X+ISO1) from A4 X ISO1 F1 females (drosophila).
We had some illumina data for A4 and ISO1 strains with which we did trio binning.The trio binning worked quite well.
For 24 kb library the initial depth was around : 184X. And the binning procedure binned A4 - 90.93X and ISO1 - 91.44X For 19 kb library the initial depth was around : 219X . And the binning procedure binned A4 - 107.2 X and ISO1 - 107.07X
I had started the work few weeks back and I used the branch Canu branch v1.9 +366 changes (r9576 7f9aa6485e4ecbed6fa12fa593aad12303650acc) . As per the nature biotech paper I used the following parameters .(I did trimming as well assembly)
1) The 24 kb library did fantastic for both haplotypes (barring two or three mis-assemblies) but it gave us an unexpected genome size of ~174 Mb. Usually for drosphila we get a genome size of 145 Mb or so with the Pacbio CLR or Nanopore. The preliminary results (for now) show that the extra 25-30 Mb are pericentromeric sequences (which the previous assemblies used to collapse). That is great news.I am attaching the report for the 24 kb library for both haplotypes.
2) The 19 kb library was showing some very strange behavior. Though the binning worked well the assemblies were off ( NG50 ~ 1.12 Mb and genome size of ~268 Mb - this can never happen probably).Also when I looked at [UNITIGGING/OVERLAPS] the unique overlaps were only ~ 41 %. I am attaching reports for both haplotypes.
( I repeated the assembly again assuming I did some mistake while running it initially. But I got the same results . I also assembled the 11 kb library but since it had around ~19X per haplotype, I was not expecting great results. But it gave me quite a decent assembly given the low depth (NG50 of ~5Mb). The error rate of 19 kb library is somewhere in between 11kb & 24 kb and the assembly worked well for both 11kb and 24 kb with correctedErrorRate=0.015 and is not working for 19kb library. Any reason why?. Also I mapped the haplotype seperated reads of 19kb library to the the A4 and ISO1 current reference and I saw no bias in coverage (uniform coverage across all major chromosomal arms) )
Also since the official release of Canu 1.9 (a week back) the -pacbio-hifi option started getting supported (Previously I did not try it since I read in one of the issues that it was still experimental). So I decided to give it a try on both 19kb and 24kb libraries. I only did the ISO1 haplotype for testing. The binning statistics were same as the previous assemblies
Version : Canu 1.9 release
For 24 kb library the stats are as follows
The report : 24kb_hifi_par.ISO1A4-haplotype_ISO1.report.txt
For 19kb libraray the stats are as follows
The report : 19kb_hifi_par_ISO1A4-haplotype_ISO1.report.txt
Now it seems the 19kb library is working and it gave us a slightly better assembly as compared to 24 kb one.(Though the NG50 looks better as compared to --pacbio-corrected one . When I looked at the biggest contig ((40 Mb and 31 Mb) ) in both of them , it had joined two big chunks of different chromosomal arms. Also the genome size we are getting is a little less than the previous --pacbio-corrected assemblies). The coverage shown in the canu report is less than the input coverage but I am guessing that is because the reads are homo-polymer compressed. The error rate used for --pacbio-hifi assembly was correctedErrorRate=0.025 . Any particular reason for increasing it from 0.015.