Tweet
How should I map transcript data in the absence of a good reference genome sequence?
I'm playing around with NGS data for an organism that has a fairly bad reference genome sequence (Schmidtea mediterranea, FWIW). In order to work out how bad, I split the reference sequence up to 100bp sequences, filtered out low-complexity sequences, then remapped back to the original sequence using bowtie2, and found quite a few places that mapped to >20 different contigs (~2000 by my rough guess using uniq/sort). This means that I'm not particularly comfortable using this genome for mapping our RNAseq data.
We now have heaps of RNAseq data and would like to look at differential expression and splice variants. We have some reference transcriptome sequences to map to, but due to some variation in reported sequences, can't be sure if two contigs from the transcriptome are different gene copies, or different isoforms of the same gene. So I have a few questions about the analysis:
Would it be better to create a new reference transcriptome from our data (about 450M reads, 35/50 paired-end on a SOLiD4, expected transcriptome size is about 20-30Mb), or use the previously published transcriptomes?
I like what cufflinks can do in identifying isoforms, but aren't sure how it responds to transcriptome mappings. Can it combine multiple contigs and notice that they are actually the same gene?
Is there any point in using tophat on a transcriptome, given that there shouldn't be any large breaks in the reference transcript sequences?
Other comments / questions would be appreciated, because this will be the first "real" analysis that I have done.
I'm playing around with NGS data for an organism that has a fairly bad reference genome sequence (Schmidtea mediterranea, FWIW). In order to work out how bad, I split the reference sequence up to 100bp sequences, filtered out low-complexity sequences, then remapped back to the original sequence using bowtie2, and found quite a few places that mapped to >20 different contigs (~2000 by my rough guess using uniq/sort). This means that I'm not particularly comfortable using this genome for mapping our RNAseq data.
We now have heaps of RNAseq data and would like to look at differential expression and splice variants. We have some reference transcriptome sequences to map to, but due to some variation in reported sequences, can't be sure if two contigs from the transcriptome are different gene copies, or different isoforms of the same gene. So I have a few questions about the analysis:
Would it be better to create a new reference transcriptome from our data (about 450M reads, 35/50 paired-end on a SOLiD4, expected transcriptome size is about 20-30Mb), or use the previously published transcriptomes?
I like what cufflinks can do in identifying isoforms, but aren't sure how it responds to transcriptome mappings. Can it combine multiple contigs and notice that they are actually the same gene?
Is there any point in using tophat on a transcriptome, given that there shouldn't be any large breaks in the reference transcript sequences?
Other comments / questions would be appreciated, because this will be the first "real" analysis that I have done.
Comment