Comparing RNA-seq data with different read lengths
Hi All,
I work with SOLiD RNA-seq data. I have sequencing from the version 4 machine and sequencing from the 5500xl. These have different reads lengths 50bp and 75bp, respectively.
Because these reads are different lengths I assume they will map differently. So I was wondering whether you can compare expression if you have samples from the v4 and 5500xl? How would you take into account biases from read length?
I have tried to test the affect of read length on mapping/ raw counts produced. Using the 5500xl data only, I compared 75bp to the same samples artificially trimmed to 50bp. I get more reads mapping in the trimmed 50bp, this doesn't surprise me as the quality will be higher (25bp from the 3' are trimmed). Though by altering the max quality for mismatched base pairs i can get mapping within 0.24% e.g. trimmed 62.76% mapping, untrimmed 62.52% mapping.
I then use HTseq to count the number of reads mapping to genes. I use edgeR to assess differential expression between untrimmed and trimmed reads (removing low expressed reads). There are approx 110 genes differentially expressed between trimmed and untrimmed. This differential expression worries me because if even within the same samples you can get differential expression due to read length changes then how can i compare data generated on the version 4 with the 5500xl? Or how, for that matter, could anyone ever compare data from other platforms with out introducing biases from read length variation?
I have been searching to see if anyone else has looked at biases introduced from read length but I cant find anything - everyone has focused on gene length, GC content etc.
Any thoughts and/or help would be appreciated. Thanks.
I work with SOLiD RNA-seq data. I have sequencing from the version 4 machine and sequencing from the 5500xl. These have different reads lengths 50bp and 75bp, respectively.
Because these reads are different lengths I assume they will map differently. So I was wondering whether you can compare expression if you have samples from the v4 and 5500xl? How would you take into account biases from read length?
I have tried to test the affect of read length on mapping/ raw counts produced. Using the 5500xl data only, I compared 75bp to the same samples artificially trimmed to 50bp. I get more reads mapping in the trimmed 50bp, this doesn't surprise me as the quality will be higher (25bp from the 3' are trimmed). Though by altering the max quality for mismatched base pairs i can get mapping within 0.24% e.g. trimmed 62.76% mapping, untrimmed 62.52% mapping.
I then use HTseq to count the number of reads mapping to genes. I use edgeR to assess differential expression between untrimmed and trimmed reads (removing low expressed reads). There are approx 110 genes differentially expressed between trimmed and untrimmed. This differential expression worries me because if even within the same samples you can get differential expression due to read length changes then how can i compare data generated on the version 4 with the 5500xl? Or how, for that matter, could anyone ever compare data from other platforms with out introducing biases from read length variation?
I have been searching to see if anyone else has looked at biases introduced from read length but I cant find anything - everyone has focused on gene length, GC content etc.
Any thoughts and/or help would be appreciated. Thanks.