NGMLR very slow on bovine nanopore reads

[sdjebali]

Dear all,

First of all, thanks for this very nice development.

I just wanted to report the fact that on some quite heavy ONT runs from bovine, NGMLR followed by sort was very slow (about 4 days for 4 million reads).

And I was wondering if I was using the tool correctly (right parameters)?

I tried with the first 1 million reads like this:
zcat $fastq | head -n 4000000 | ngmlr --presets ont -t 22 -r $genome
| samtools sort -@ 6 -o $output
and it took 5h23 to complete

I then tried with the second 1 million reads like this:
zcat $fastq | tail -n+4000000 | head -n 4000000 | ngmlr --presets ont -t
22 -r $genome | samtools sort -@ 4 -o $output
and it took 24h10 to complete

I am using NGMLR version 0.2.8 and samtools version 1.9, and here are the details about my machine :
Linux tatum 4.19.0-5-amd64 #1 SMP Debian 4.19.37-5+deb10u2 (2019-08-08)
x86_64 GNU/
24 processors
Linuxprocessor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 45
model name : Intel(R) Xeon(R) CPU E5-4610 0 @ 2.40GHz

Any advice would be warmly welcome?

Best,
Sarah

[fritzsedlazeck]

Thanks Sarah,
do you have an average read length? Its likely but unfortunate that some of your 2nd patch reads are very long..
Thanks
Fritz

[sdjebali]

Indeed there seems to be a big read length difference between the two batches.

I ran Nanoplot on them and here are the results :

• First 1 Million reads:
  General summary:
  Mean read length: 4,722.5
  Mean read quality: 4.4
  Median read length: 906.0
  Median read quality: 4.2
  Number of reads: 1,000,000.0
  Read length N50: 14,404.0
  Total bases: 4,722,479,679.0
  Number, percentage and megabases of reads above quality cutoffs

Q5: 367454 (36.7%) 3015.3Mb
Q7: 8 (0.0%) 0.1Mb
Q10: 0 (0.0%) 0.0Mb
Q12: 0 (0.0%) 0.0Mb
Q15: 0 (0.0%) 0.0Mb
Top 5 highest mean basecall quality scores and their read lengths
1: 7.0 (17272)
2: 7.0 (9848)
3: 7.0 (25242)
4: 7.0 (12091)
5: 7.0 (25093)
Top 5 longest reads and their mean basecall quality score
1: 2210466 (3.6)
2: 1850945 (3.8)
3: 1772717 (3.6)
4: 1685671 (3.9)
5: 1563326 (3.9)

• second 1 Million reads
  General summary:
  Mean read length: 13,668.0
  Mean read quality: 11.1
  Median read length: 13,451.0
  Median read quality: 11.8
  Number of reads: 1,000,000.0
  Read length N50: 16,657.0
  Total bases: 13,668,019,254.0
  Number, percentage and megabases of reads above quality cutoffs

Q5: 963153 (96.3%) 13574.0Mb
Q7: 937982 (93.8%) 13387.4Mb
Q10: 781757 (78.2%) 10950.3Mb
Q12: 446035 (44.6%) 6333.8Mb
Q15: 165 (0.0%) 1.6Mb
Top 5 highest mean basecall quality scores and their read lengths
1: 16.3 (2090)
2: 16.2 (243)
3: 16.1 (362)
4: 16.1 (570)
5: 16.1 (1509)
Top 5 longest reads and their mean basecall quality score
1: 884004 (3.7)
2: 274368 (5.2)
3: 187850 (4.8)
4: 150969 (3.8)
5: 124444 (9.8)

so 13kb vs 4kb

If we still want to use NGMLR on these data, is there any option that can speed the process up?

Best,
Sarah