question about usage of --min_range

[ZheFrench]

Hi folks,

Don't understand well how to use this parameter. From doc : --min_range option, to provide higher (positive values) or lower (negative values) stringency. --min_range is the minimum value any individual pair's ΔPSI can be.

So you can set up negative value. For example , min_range =-1, and so you expecting to have no filtering on PSI for replicate.

For paired comparison, setting min_range=5 and min_dPSI=20 means you filter out dpsi < 20 and psi should be > 5 at least in each replicates of your pairs ?

Thanks

[mirimia]

Hey,

Summarized in this way, it certainly looks confusing. I will try to improve the explanation in the README. Here it's the idea:

• For NON-PAIRED comparisons, --min_range is the minimum difference that can be between the two PSI distributions. So, if you have for group A PSI_A1=10 and PSI_A2=20, and for group B PSI_B1=30 and PSI_B2=50, the dPSI would be 40-15 = 25, and the "range difference" 30-20 = 10. I.e. the difference between the maximum value of group A and the minimum of group B.

• For PAIRED comparisons, it is just the minimum paired dPSI. So, in the above example: dPSI_1 = 30-10 = 20, and dPSI_2= 50-20 = 30. So the minimum paired dPSI is 20 (dPSI_1).

By using positive --min_range values, you make sure that the PSI distributions of group A and B do NOT overalap. On the contrary, use of negative --min_range allows PSI distributions to overlap. So, if, instead, group A PSI_A1=10 and PSI_A2=30, and for group B PSI_B1=20 and PSI_B2=50, the "range difference" here would 20-30 = -10.

I hope it is clearer now.

[ZheFrench]

For PAIRED comparisons, if I say it differently, min_range is the difference at the level of each paired replicate value whereas min_diff is the difference using the average of the replicates. So I can keep my threshold with min_diff but be less restrictive with a lower min_range for example. Your change in the doc is cleaner by the way...:)

It's better but still get confuse with positive/negative values for min range. I don't understand the concept of overlapping distribution.

_So, if, instead, group A PSI_A1=10 and PSI_A2=30, and for group B PSI_B1=20 and PSIB2=50, the "range difference" here would 20-30 = -10.

Ok so for positive it would be 30 - 20 = 10 (max in A, min in B). What's the goal /idea behind this ?

[mirimia]

I'm not sure I understand. Do you mean of allowing overlapping distributions? Sometimes if you have many replicates it may be wise to do it. Although if you have many replicates I'd use "regular" stats...

[ZheFrench]

I have 3 replicates in each condition... Humm I do not understand the sign of min_range. I think i understand that is the minimum paired dPSI Test Control Rep1 5 10 Rep2 5 7 Rep3 10 7

Here Rep2 |5-7|=2 ...if you set minrange= 1, you will filter out this event... What happen if you set -1 ?

[mirimia]

Here you will have (for paired analysis):

dPSI_1 = 5-10 = -5 dPSI_2 = 5-7 = -2 dPSI_3 = 10-7 = 3

av_dPSI= (-5+-2+3)/3= -1.333

So, if you had --min_dPSI 1 --min_range 1 it won't make it, because the individual dPSIs are of different signs...

you'd need something like --min_range -3 or so. But it's obviously a case that should be filtered out! :-)