[miso-users] question about inclusion ratio and PSI

[Yarden Katz]

[Note: I am traveling and am slower than usual in responding]

Hi, 

See comments below.



On Aug 13, 2013, at 1:55 AM, yu_peng wrote:

> Hi, colleague:
>  
> I have several questions:
> 1. In your 2008 Nature paper,the 'inclusion ratio' is defined as the ratio of the number of 'inclusion’ reads to inclusion plus ‘exclusion’ reads.
> Are the reads used here those assigned by MISO using Bayesian method? 

That ratio from Wang et. al. 2008 is referred to in the MISO paper as "PsiSJ".  PsiSJ is a simple analytic estimator that takes the inclusion densities you refer to and gives you an estimate of Psi (with no confidence intervals).  MISO does not reduce to a simple estimate like this and instead infers a distribution over the Psi values.

MISO considers a range of possible assignments of reads to isoforms (in proportion to their probability) and uses that range to calculate Psi values.  So there's never a singular assignment of reads to isoforms that is used to calculate PsiSJ in MISO.

>  
> 2. When calculating PSI, the density of reads were used (reads per position), i.e. one need to divide the above (assigned) reads with some length.
> However, I have not found the definition of the effective length in the paper. In the supplementary file, it was noted that 28nt were kept from both
> sides of a junction (cause your original read is 32bp, thus can assure at least 4bp overlapping the next exon). So in my understanding, the effective
> length of a junction is 28+28 (this could be changed to real read length minus 4 ), while for event including a middle exon, the length should be 28+28+length of included exon. Am I right?

To reiterate, MISO Psi values are not calculated with these densities.  But assuming you wanted to calculate the 2008 PsiSJ estimate, you would use the lengths of isoforms to calculate the number of possible positions that a read can land in.

If you have an exon of length L and reads of length K, there are L - K + 1 positions.  Across junctions, you have to take into account overhang as you write, and you can do that using the fact that an overhang constraint of O will give you: K + 1 -2*O positions for read length K.  So yes, the overhang limits the number of possible read starting positions across junctions but not within exons.


> And for A5SS or A3SS events, as there are reads spanning the boundary of two events, how do you deal with them?

It's exactly the same as skipped exons to deal with number of possible read positions etc. if that's what you're asking.

>  
> 3. Currently I am analyzing a set of pair-end RNA-sequencing data. The insertion size is about 150nt but the read length is 101, so two reads overlap.
> When using MISO with pair-end mode, it showed low read coverage for many events. Is MISO vulnerable for such condition?

Are you using short event annotations (two-isoforms) or full whole isoform annotations?  Read pairs could be thrown out if they are outside the boundary of the event.  

Are you use that you fed --paired-end the right insert length mean and standard deviation?  MISO should be able to deal with 'negative' insert lengths, i.e. cases where the read pairs overlap.  
>  
> 4. And the last question. In my result of TandemUTR analysis, it is seen that the assigned read are greatly different between events, but the PSIs are
> similar (see below), is this normal?
>  
> event_name      miso_posterior_mean     ci_low  ci_high isoforms        counts  assigned_counts chrom   strand  mRNA_starts     mRNA_ends
> 2d  0.28,0.26,0.27,0.20     0.24,0.22,0.25,0.18     0.31,0.31,0.33,0.22     '2d.1.0','2d.2.0','2d.3.0','2d.4.0'     (0,0,0,0):1163,(0,0,0,1):6,(0,0,1,0
> ):1,(1,0,0,0):203       0:203,1:0,2:1,3:6       chr3    +       126545676,126543284,126518921,126303617 126546040,126543378,126521721,126304424
>  
This looks like the output for one event -- so I can't tell what I should compare it to?

Best, --Yarden

> Thanks.
>  
> Peng  
>  
> ***************************************
> Peng Yu, Ph.D.
> Institute of Molecular Medicine
> Room 303, Yingjie Center
> Peking University
> Beijing 100871, China
> ****************************************
>  
>