Median TIN is anticorrelated with RIN

[Gonzalo S. Nido]

Hi,

I used tin.py to calculate the median TIN of my RNAseq samples and the results are the opposite to what expected: the TIN is significantly anticorrelated with RIN values (and, of course, with insert size and DV200 score). According to your paper, it should be positively correlated.

Regards,

Gonzalo

[Liguo Wang]

Hi,

TIN is a measurement of "reads coverage uniformity" across gene body, assuming that RNA-seq data generated from low-quality sample has bad "coverage uniformity" due to degradation. However, some special RNA-seq protocols can still generate good reads coverage from degraded RNA sample (such as exome capture RNA-seq protocol: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4561495/). In this case, you will get high median TIN score even the RIN is low. 

That being said, RIN measures if the RNAs have been broken-down while TIN measures if the reads coverage has been distorted due to RNA broke down. They are not always positively correlated, but I don't expect they have a negative correlation.

The median TIN scores for your samples are very high (most > 70), suggesting good quality.

-Liguo

--
You received this message because you are subscribed to the Google Groups "rseqc-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rseqc-discuss+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gonzalo S. Nido]

First of all, thank you for your prompt response.

The data I'm analyzing is ribo-depleted RNA-seq generated from post-mortem tissue. Some of my samples are of bad quality (in terms of RNA integrity, with RINs ranging from 3 to 9) because they were frozen after many hours post-mortem. As expected, the longer the period after death, the worse the RNA quality is (measured in RIN, average insert size, or DV200). But median TIN, however, is the opposite, and anticorrelates significantly with all those measures.

With RIN in particular, it has a correlation coefficient of -0.67 with a p-value of 1.40e-07. This is the scatterplot with a linear regression on top:

To unsubscribe from this group and stop receiving emails from it, send an email to rseqc-discus...@googlegroups.com.