Segmentation Fault when Mapping to Human chrX Only (Tried different --genomeSAindexNbases values)

[Dadi Gao]

Hi,

I'm currently using STAR 2.4.2a. In my past experience, STAR always finishes alignment successfully, until now I'm trying to map reads to human X chromosome only. In my first round, my original fastq files were mapped to the whole human genome index and only unique maps were allowed. All the samples finished successfully. For unmapped reads in each sample, I tried to re-map them against the X chromosome only and up-to 20 hits were allowed.

In order to carry out the second round, I built STAR genome index for human X chromosome, after calculating min(14, log2(GenomeLength)/2 - 1). As X chromosome is 155273187 bp, this value ended up as 13.

STAR --runMode genomeGenerate \

--genomeDir star_GRCh37.75_Xonly \

--genomeFastaFiles X.fa \

--sjdbGTFfile X.gtf \

--sjdbOverhang 150 \

--genomeSAindexNbases 13

This completed without problem. Then I run the mapping step:

STAR --runThreadN 8 \

--genomeDir star_GRCh37.75_Xonly \

--twopassMode None \

--outFilterMultimapNmax 20 \

--outFilterMismatchNoverLmax 0.05 \

--outSAMtype BAM Unsorted \

--outReadsUnmapped Fastx \

--alignIntronMin 21 \

--alignIntronMax 0 \

--alignEndsType EndToEnd \

--outFileNamePrefix Remap. \

--readFilesIn read1.fq read2.fq

It turned out: 

Sep 22 11:51:05 ..... Started STAR run

Sep 22 11:51:06 ..... Loading genome

Sep 22 11:51:11 ..... Started mapping

Segmentation fault

I also tried to use different genomeSAindexNbases values, from 2 to 14, to rebuild the genome index, but all failed at the mapping stage with segmentation fault.

I attached my Log.out file here. I found a lot of warnings like "not enough space allocated for transcript. Did not process all windows for read...". This never occurred when I mapped the same reads files to the whole human genome reference. (BTW, the reads are 150bp paired-end before adapter trimming)

Any help or suggestion is really appreciated.

UPDATE: I complied a newer version 2.5.2a and re-run the alignment against the X chromosome index built under 2.4.2a. It finished without any error. It turns out that the seg-fault was due to the aligner itself instead of index. I'm wondering what has been changed between the two versions?  Do I need to re-do the first round of mapping, which is to map all the original fastq files (from 200 samples) against the whole human genome, using 2.5.2a?

Best,

Dadi

[Alexander Dobin]

Hi Dadi,

a number of changes were made from 2.4.2a to 2.5.2b. The change in the read alignments should be negligible, but I would check it (i.e. run two versions for a few samples) to see the magnitude of the changes.

Why are you re-mapping the reads to the X-chromosome only? Which parameters did you change compared to the whole genome mapping?

Cheers

Alex