[bruce01]

Dpryan, yes have tried using wildcards as input to test it works, I get a segmentation fault. When I run it with all filenames included as standard it runs fine. I have a lot of samples, with variable numbers of fastq files per sample, and want a single script to submit to a queue. So inputting all fastq by hand is not an option, hence my original question.

Concatenating the fastqs will mean I have to uncompress them, using computing time and I am keen to go from the .gz that my facility have supplied. This can't be too big of a problem is it?

[dpryan]

I have made files containing fastq in the specified format (fq_r1_1,..,fq_r1_n). Can I use these when submitting the STAR job? Ie:

STAR [options] readFilesIn $files/file_read1 $files/file_read2

Have you just tried the following?

STAR --readFilesIn Sample1_r1_1.fq,Sample1_r1_2.fq,Sample1_r1_3.fq... Sample1_r2_1.fq,Sample1_r2_2.fq,Sample1_r2_3.fq...

You could also just concatenate the files together as appropriate and use the result.

[bruce01]

Hi all, sorry for the basic question:

I am writing a bash script to submit star jobs, remove duplicates, get counts etc. The dataset I have has multiple fastq per sample, but different numbers for each. I have made files containing fastq in the specified format (fq_r1_1,..,fq_r1_n). Can I use these when submitting the STAR job? Ie:


STAR [options] readFilesIn $files/file_read1 $files/file_read2

?

Have tried a few ways to do this but can't figure it out or get STAR to accept input. I am a 'midrange' bioinformatics PhD, so don't hold back on most efficient or crazy way of doing this!

Thanks in advance,

Bruce.

[[email protected]]

As far as I know TopHat does not produce a GTF file on its own, at least it was true for the last version I tried (~2.0.3). You need to feed the alignments to Cufflinks, which will assemble and quantify transcripts, and produce the GTF file.

You are right, sorry I mixed it up a bit. Thanks for the information on the options I should use.

[alexdobin]

I am pretty new to RNA-seq analysis and I am now using STAR instead of Tophat and I am very satisfied with both the quality of the results and the speed at which I get them. One thing I miss though is the .GTF file I get from Tophat that contains new genes predicted based on the reads and splice junktions.
Does anyone know if there is a way I can combine an existing GTF file with the .tab file to create a new .GTF (or GFF) file containing newly predicted gene sites (with random names for these)?

As far as I know TopHat does not produce a GTF file on its own, at least it was true for the last version I tried (~2.0.3). You need to feed the alignments to Cufflinks, which will assemble and quantify transcripts, and produce the GTF file.

You can run Cufflinks on STAR alignments.
If you have un-stranded RNA-seq data you will need to run STAR with --outSAMstrandField intronMotif option, which will generate the XS strand attribute for all alignments that contain splice junctions. The spliced alignments that have undefined strand (i.e. containing only non-canonical junctions) will be suppressed.

If you have stranded RNA-seq data, you do not need to use any specific STAR options. Instead, you need to run Cufflinks with the library option --library-type options. For example,
cufflinks ... ... --library-type fr-firststrand
should be used for the “standard” dUTP protocol. This option has to be used only for Cufflinks runs and not for STAR runs.
It is recommended to remove the non-canonical junctions for Cufflinks runs using STAR's options:
--outFilterIntronMotifs RemoveNoncanonical OR RemoveNoncanonicalUnannotated

[[email protected]]

I am pretty new to RNA-seq analysis and I am now using STAR instead of Tophat and I am very satisfied with both the quality of the results and the speed at which I get them. One thing I miss though is the .GTF file I get from Tophat that contains new genes predicted based on the reads and splice junktions.
Does anyone know if there is a way I can combine an existing GTF file with the .tab file to create a new .GTF (or GFF) file containing newly predicted gene sites (with random names for these)?

[alexdobin]

Alex, I successfully used STAR to generate the SAM file. But I can't find how to specify the output for chimeric alignments. Should I use "--outSAMunmapped Within" to include everything in the SAM and use samtools to find chimeric alignments? And also for "--outReadsUnmapped", does it include chimeric and singleton?

Thanks.

To switch on chimeric detection and output, you would need to specify non-zero --chimSegmentMin, which is a minimum length of a segment (piece) of which chimeras are made. For example, if you have 2x100 PE reads and specify --chimSegmentMin, you could have a chimera in which one segment of (100-mate1+80-mate2) bases maps non-chimerically to one chromosome, and another segement of 20b-mate2 maps to another chromosome.
The Chimeric output will go into Chimeric.out.sam and Chimeric.out.junction files.

Note that the same read can have both acceptable non-chimeric (output to Aligned.out.sam) and chimeric alignments (output to Chimeric.out.*). A read is considered "unmapped" if it does not have an acceptable non-chimeric alignment, and --outSAMunmapped Within will output "unmapped" reads into Aligned.out.sam without alignment coordinates (which allows to fully reconstruct fastq file from the SAM file), while --outReadsUnmapped Fastx will output them into a fastq or fasta files.

There are other parameters that control chimeric detection:
chimJunctionOverhangMin 20
int>0: minimum overhang for a chimeric junction
chimScoreMin 0
int>0: minimum total (summed) score of the chimeric segments
chimScoreDropMax 20
int>0: max drop (difference) of chimeric score (the sum of scores of all chimeric segements) from the read length
chimScoreSeparation 10
int>0: minimum difference (separation) between the best chimeric score and the next one
chimScoreJunctionNonGTAG -1
int: penalty for a non-GT/AG chimeric junction

[Auction]

You will need to generate special genome files for STAR.
This is done with the following command:
STAR --runMode genomeGenerate --genomeDir /path/to/GenomeDir --genomeFastaFiles /path/to/genome/fasta1 /path/to/genome/fasta2 --runThreadN <Nthreads>
If you want to use annotations for improved mapping accuracy, you also need to use:
--sjdbGTFfile /path/to/Annot.gtf --sjdbOverhang <N>, where ideally N=ReadMateLength-1, or you could generically use ~100.

Alex, I successfully used STAR to generate the SAM file. But I can't find how to specify the output for chimeric alignments. Should I use "--outSAMunmapped Within" to include everything in the SAM and use samtools to find chimeric alignments? And also for "--outReadsUnmapped", does it include chimeric and singleton?

Thanks.

[alexdobin]

Hey, does anyone know if you need the reference genome indexed according to Star because I know for tophat2 the reference genome needs to be indexed *.b2t (bowtie2)

Thanks,
Nino

You will need to generate special genome files for STAR.
This is done with the following command:
STAR --runMode genomeGenerate --genomeDir /path/to/GenomeDir --genomeFastaFiles /path/to/genome/fasta1 /path/to/genome/fasta2 --runThreadN <Nthreads>
If you want to use annotations for improved mapping accuracy, you also need to use:
--sjdbGTFfile /path/to/Annot.gtf --sjdbOverhang <N>, where ideally N=ReadMateLength-1, or you could generically use ~100.

[alexdobin]

For loci that vary by one base substitution, what would be the best parameters to discriminate the two alleles?

Also, I want to tailor the output for single best match, and if two are equivalent, no matches reported.

Regards,

-Tom

Sorry, did not check this thread for awhile, just found your question.

With default parameters, if two alignments differ by one mismatch, only the best one would be reported. This is controlled by --outFilterMultimapScoreRange, (=1 by default) which defines the range of alignment scores that are reported as multi-mappers.

The max number of alignments that allowed for output is controlled by --outFilterMultimapNmax. It's 10 be default, so any read with 10 or fewer alignments with scores >= BestScore-1 will be reported.

[alexdobin]

Interesting, thanks. Unfortunately for mixed-species mapping (At least when one has alternative splicing and the other does not) the way tophat works it's essentially the same as running bowtie for bacterial reads first (Though I think there might be a way around it). There's also problems with the max size of a bowtie index :/ Would the size be a concern for STAR? Or could it run a database of anysize? (Hopefully RAM shouldn't be too much of an issue for me)

Also just to clarify, you don't know STAR would perform with bacterial reads?

Sorry, did not check this thread for awhile, just found your questions.

STAR will work with database of any size, however, required RAM scales linearly with database size, at ~10*DBsize bytes.

STAR works fine with bacterial reads, this is the work we did with long and small RNA-seq for some bacteria.

[Nino]

Hey, does anyone know if you need the reference genome indexed according to Star because I know for tophat2 the reference genome needs to be indexed *.b2t (bowtie2)

Thanks,
Nino

[thomasblomquist]

For loci that vary by one base substitution, what would be the best parameters to discriminate the two alleles?

Also, I want to tailor the output for single best match, and if two are equivalent, no matches reported.

Regards,

-Tom

[bob-loblaw]

We used STAR in two mixed-species settings: human + viruses (~4,500 viruses from NCBI) and human + mouse, and in both cases it worked well as far as we could tell. You have to keep an eye on RAM, since STAR would need ~10*TotalGenomeSize bytes of RAM (~50GB for human+mouse), and if you have a large number of small chromosomes/scaffolds/contigs, you would need to reduce --genomeChrBinNbits as I explained in the previous post.

For mixed-species mapping I strongly agree with other users who recommended mapping to a combined genome rather than a 2-step mapping. When a mapper aligns a read to combined genome it considers various possible alignments and picks the best one. This reduces the false positives and negatives for unique calls to each species. On the other hand, in the 2-step method you force alignments to one of the species in the 1st step, so you results will be strongly biased towards the 1st species.

Interesting, thanks. Unfortunately for mixed-species mapping (At least when one has alternative splicing and the other does not) the way tophat works it's essentially the same as running bowtie for bacterial reads first (Though I think there might be a way around it). There's also problems with the max size of a bowtie index :/ Would the size be a concern for STAR? Or could it run a database of anysize? (Hopefully RAM shouldn't be too much of an issue for me)

Also just to clarify, you don't know STAR would perform with bacterial reads?

[thomasblomquist]

Had the chance to test your recommended parameters for my specific needs. This is not my normal mode of operation, but... HOLY POOP!!! I cut a 2 hour BFAST match and sorting into bins down to 10 seconds!!! And the specificity and sensitivity is off the hook! When I do some more formal testing I will provide back metrics.

You are a computer jedi master! This will make analysis of clinical sequencing data a breeze. Very intuitive command lines.

-Tom Blomquist