Indexing 34Gbp Reference Genomes With Bwa -A Bwtsw

Question

[Ahdf-Lell-Kocks]

I have tried to index 34GB worth of primate genomes in one single go using bwa -a bwtsw and it went all the way into doing that but then crashed at this point:

[BWTIncConstructFromPacked] 7090 iterations done. 70465272750 characters processed.
[BWTIncConstructFromPacked] 7100 iterations done. 70493213214 characters processed.
[BWTIncConstructFromPacked] 7110 iterations done. 70518042750 characters processed.
[bwa_index] 118525.77 seconds elapse.
[bwa_index] Update BWT... /tools/lsf/spool/1327577148.2686228: line 8:  7412 Segmentation fault      (core dumped) ~/src/bwa/bwa-0.6.1/bwa index -a bwtsw primates.fa

I tried this on a 30000M job, trying again now (still chugging along) on 60000M.

Any suggestions on what might be happening here?

[Michael]

Can you give a bit more information, especially RAM size might be important, because I have seen bwa segfault because of memory, at least it wouldn't tell: 'couldn't aquire 200G of memory', no no...

Answer 1

[Pierre Lindenbaum]

You can try to find the error using gdb.

in $BWA/Makefile check that the -g option is set for CFLAGS

CFLAGS=         -g -Wall -m32

clean and re-compile:

make clean
make

run gdb and call backtrace after the failure

$ gdb /path/to/bwa 
(gdb) run -a bwtsw primates.fa
(...)
Error
(...)
(gdb) backtrace

It should give us an insight about the problem.

Answer 2

[Ahdf-Lell-Kocks]

Answering my own question, it needed about 60Gb of memory to do the indexing, it now works, I tried bwa bwasw -Z 10 and will load the index in about 64Gb and align normally.

[lh3]

I guess this is the first time that someone has run BWA on a 34GB genome. Yes, bwa/bwa-sw needs about twice memory as your reference genome.

[Evgeniia Golovina]

Excuse me, twice memory is required only for indexing or for mapping as well?

[lh3]

I guess this is the first time that someone has run BWA on a 34GB genome. Yes, bwa/bwa-sw needs about twice memory as your reference genome.