I have run several different data sets through the Tuxedo pipeline, but have just noticed that the isoforms.fpkm_tracking files are all messed up in the FPKM columns. According to the Cufflinks manpage, columns 10-12 of all the tracking files are defined as follows:
10 q0_FPKM 8.01089 FPKM of the object in sample 0
11 q0_FPKM_lo 7.03583 the lower bound of the 95% confidence interval on the FPKM of the object in sample 0
12 q0_FPKM_hi 8.98595 the upper bound of the 95% confidence interval on the FPKM of the object in sample 0
However, in my files, MANY column 10 FPKM values are several orders of magnitude higher than column 12, FKPM_conf_hi, which is the upper bound. Has anyone else noticed this in their files? I cannot find any explanation for this, and like I said, all of these files, even from different datasets and different experiments, show the same phenomenon.
I am using TopHat v2.0.9, cufflinks v2.1.1.
I have attached a pdf image of a scatterplot to visualize these differences, because it seems that this error is correlated to fragment length. On the y-axis is the difference FKPM(column 10 above)-FPKM_conf_hi(column 12, upper bound). This SHOULD be negative. All points above the x axis are erroneous.
The x-axis is fragment length (Column 8 of isoforms.fpkm_tracking). You can see that this error seldom occurs at higher fragment lengths.
Please help!
10 q0_FPKM 8.01089 FPKM of the object in sample 0
11 q0_FPKM_lo 7.03583 the lower bound of the 95% confidence interval on the FPKM of the object in sample 0
12 q0_FPKM_hi 8.98595 the upper bound of the 95% confidence interval on the FPKM of the object in sample 0
However, in my files, MANY column 10 FPKM values are several orders of magnitude higher than column 12, FKPM_conf_hi, which is the upper bound. Has anyone else noticed this in their files? I cannot find any explanation for this, and like I said, all of these files, even from different datasets and different experiments, show the same phenomenon.
I am using TopHat v2.0.9, cufflinks v2.1.1.
I have attached a pdf image of a scatterplot to visualize these differences, because it seems that this error is correlated to fragment length. On the y-axis is the difference FKPM(column 10 above)-FPKM_conf_hi(column 12, upper bound). This SHOULD be negative. All points above the x axis are erroneous.
The x-axis is fragment length (Column 8 of isoforms.fpkm_tracking). You can see that this error seldom occurs at higher fragment lengths.
Please help!
Comment