I agree with Piazza. It is fairly common to see fastq files are in different sizes, no matter whether they are from a same batch of sequencing or not, and it usually does not affect downstream analysis at all. However it raises suspicions of sequencing quality when low count fastqs are of the same "treatment". It matters a lot to your analysis if the difference is indeed a result of the poor quality of your original RNA (it looks to me the bad quality of your RNAs is the issue, as the DNA-array revealed something similarly). I believe this can be revealed by looking at the percentages of your reads mapped to the transcriptome, and compare them to the fastqs with more reads.  If the percentages are similar I will make a note of the low count issue and carry on to the analysis. As of today, it is still the best to use sequencing companies, as sequencing does require a lot of optimizations and experiences, although you will lose a bit of control of sample handling. 

No, what you are seeing is probably the result of the preparation, not gene regulation in the sample. Some tissue types yield less RNA, or perhaps even lower quality RNA sure, but if you send libraries of the same quantity and size distribution you shouldn't have a difference in the number of reads returned. For instance, if you send a 10 nM concentration of a ~400 bp library, it doesn't matter if your starting material was 1 microgram or 5 micrograms of high or low quality RNA, you should still get the same number of reads returned. 

Did you use a bioanalyzer (or did the genome center) to look at your libraries before they were sequenced? Generally, the most common reasons for a low return of reads are primer dimers in samples, poor size distribution, or incorrect quantification. In addition, if you multiplexed, you may have added your samples unevenly. 

Yes, the gene number you find is related to your sample. The samples from different times or tissues show different gene number.

Thanks for all your answers. Maybe I should make it clear:

1) the reads number variation I found was the sizes of different fastq files, the fastq files after QC. I found some files were around 800MB while some were around 1000MB.

2) DNA-array can be used to analyse cDNA. I found some samples' cDNA showed overall lower signal intensity (median of expression ratio were lower) 

So I  guess some tissues didn't yield good quality RNA. I send all my samples to a company, so I didn't know the details of library construction.  I think they adjusted the concentration before library construction. 

Hi Fang,

The answer is yes, the quality of your sample will effect the QC of your reads, albeit indirectly. The higher the purity of your RNA samples, the more comparable their RNA with any selection criteria you've created (size selection, oligo dT association, etc.) Also the measurement of the RNA will be more uniform if no contaminants are present - this can effect how the nanodrop assesses your concentration, Qubits tend to be less variant in measurement, but individual technicians will swear by the system they use. Find out which measurement tool they use and what might impact/alter its function.

If you are not processing the samples yourself, then there is only so much control you have over this. The technicians will adjust the concentration of your samples prior to any sequencing if you're comparing reads quantitatively, which I assume you are.

Could the treatment of one group vary the concentration of observed reads overall?

This could be the case if the treatment impacted translational efficiency and thereby mRNA stability/processing. So consider the impact of your treatment conditions biologically and not just molecularly to determine whether what you're observing is an inconvenient, but real result.

Hope this helps,

Pegine

Some avenues of inquiry:

1) Library quality can be assessed to some degree by running FASTQC on the FASTQ files. 2) RNA degradation will be apparent as bad rRNA traces on a Bioanalyser. 3) Degraded RNA will have a bias towards 5' reads (if a whole RNA prep). The exonucleases degrade 3' to 5'. Oligo-dT priming in polyA libraries will hide this effect to some degree, but will adversely affect your results. 4) Some of the literature on running microarrays from degraded mRNA is informative and will give you some leads to inspect in the NGS data.

Hi Fang,

sure, the quality of your input RNA will affect the read number and, possibly, the quality of your analysis. It is important however, to distinguish between RAW read count, which depends on RNA quality, amount of library loaded in the NGS instrument and so on and ultimately affects the 'size' of your FastQ, and the REAL expression level of individual genes. In RNA-Seq expression is measured using RPKM (or similar units). RPKM stands for 'Reads per Kbase per Million mapped reads': with RPKM, reads mapping to each gene are normalized using gene length (bigger genes are sequenced proportionally more than small ones) and total number of reads (if experiment #1 generates 10 million reads and experiment #2 20 million, all the genes in exp #2 will be covered 2x. To neutralize this effect RPKM accounts for total number of mapped reads in the denominator of the fraction). This allows to reliably (up to some extent) compare experiments with different raw counts.

I agree with Piazza. It is fairly common to see fastq files are in different sizes, no matter whether they are from a same batch of sequencing or not, and it usually does not affect downstream analysis at all. However it raises suspicions of sequencing quality when low count fastqs are of the same "treatment". It matters a lot to your analysis if the difference is indeed a result of the poor quality of your original RNA (it looks to me the bad quality of your RNAs is the issue, as the DNA-array revealed something similarly). I believe this can be revealed by looking at the percentages of your reads mapped to the transcriptome, and compare them to the fastqs with more reads.  If the percentages are similar I will make a note of the low count issue and carry on to the analysis. As of today, it is still the best to use sequencing companies, as sequencing does require a lot of optimizations and experiences, although you will lose a bit of control of sample handling. 

I also commonly see variation between datasets in the millions of reads even though the same amount of each library was loaded onto the sequencer and they were multiplexed together. It's not a problem as long as your quality reports look similar and you have enough coverage from each library for whatever type of analysis you're doing. Usually the sequencing facility would let you know before they ran your samples if there was a major issue like large amounts of primer dimer of something else in some of your libraries that would have a large effect on your output. As Rocco says, FPKM takes into account dataset size for comparing gene expression levels.