250bp PE Miseq reads assembly k-max ?

[voutcn]

250bp PE Miseq reads assembly k-max ?

[chuckpr]

+1 I am curious about this myself as many of our 2x150 reads can be merged before assembly.

[voutcn]

is K-max 127 long enough in this case ? is there a "rule" to determine k-max ?

Honestly, I am not sure. Basically, I would not recommend setting k to larger than 100, unless the sequencing depth is very high (some single-cell data can be > 1000x). Roughly speaking, the error rate of HiSeq is about 1%, when k goes as large as 100, you may rarely see any correct kmers from reads.

However, I don't have much experience with MiSeq. So I leave this issue open here for more discussion.

can we set k-max to a higher value ? does it make any sense ?

As I know the A5-Miseq pipeline makes use of very large k-mer size for MiSeq data. So I think probably it makes sense, but you still have to try and compare before going ahead.

To use larger k value. Two files should be modified

• modify kMaxK to the value you want at https://github.com/voutcn/megahit/blob/master/definitions.h#L56
• modify the > 127 to the k-max you want at https://github.com/voutcn/megahit/blob/master/megahit#L374

[chuckpr]

Thanks, I'll give it a shot.

If I change the --k-max value in the opts.txt file and run megahit with the --continue flag can I start an assembly where a previous assembly left off? This would be a quick way to see if k > 127 helps with my data. Thanks again!

[voutcn]

@chuckpr No, you have to rerun it from the beginning. You can let k-max go as large as you want, and then evaluate the contigs assembled from different k in intermediate_contigs/k*.contigs.fa

[jvollme]

Honestly, I am not sure. Basically, I would not recommend setting k to larger than 100, unless the sequencing depth is very high

But I thought there is an iterative read correction beginning at lower kmer steps (I thought that was one of the points of doing assemblies using iterative kmer sizes)? Shouldn't that remove most such errors before you reach the high kmers?
For Miseq Data of up to 300bp l got huge improvements (concerning N50 and maximum contig length) with IDBA_UD after modifying it to allow kmer lengths upto 251 bp. I am really looking forward to trying this with megahit as megahit is far more user friendly.

[jvollme]

But I have a question regarding the exact modification for megahit:
The first change (in https://github.com/voutcn/megahit/blob/master/definitions.h#L56) is straigntforward. But what do I have to change here:

modify the > 127 to the k-max you want at https://github.com/voutcn/megahit/blob/master/megahit#L374

Can't find the value I am supposed to change here.

[voutcn]

@jvollme if you are using the latest codes, simply change the value kMaxK in definitions.h works.

[jvollme]

Thanks!
Maybe you could mention that info in your manual or readme somewehere? I think this info might be interesting for a number of people also but may be hard to find (even though Megahit is quite well documented otherwise).