Outlier detection in MAJIQ 2.1

[Rosa]

Hello, 

Has the outlier detection argument: weights, been removed from MAJIQ 2.1. ?

The version number of the MAJIQ I installed is 2.1-179b437. However, weights is no longer one of the positional arguments (positional arguments: {build,psi,deltapsi} ).

If it has been removed, is there an alternative argument to detect the outliers? Thank you!

[Yoseph Barash]

Hi Rosa,

Outlier detection is indeed not supported by the current MAJIQ 2.*. We plan to release new algorithms implemented within MAJIQ that handle better heterogeneous data and are less sensitive to outliers. We may also reintroduce the outlier detection as a MAJIQ utility tool. If you are in need of outlier detection right now I suggest you take the PSI quantification across your samples and plot those using UMAP (as described in the Norton et al and other publications). While this is not a quantitative/automated measure as we introduced in that paper it will still enable you to visually inspect your data. 

Hope this helps,

Yoseph Barash 

[Rosa]

Hi Yoseph, 

MAJIQ has been immensely helpful and thank you so much for developing the tool and continuing to improve it. 

Would you please elaborate on the method you mentioned (maybe I missed it, but I wasn't able to find UMAP related methods in Norton et al.)? My understanding is that I should run PSI on individual samples. Since the outputs all have different number of LSVs, a potential solution to that would be to combine the LSVs of a group of replicates together and construct a new matrix where all the LSVs are included. However, since each LSV often contains multiple junctions and therefore, multiple E(PSI) would be associated with one LSV, which increases the complexity of the matrix and make it a less suitable input for UMAP. I was wondering what would you recommend to use as the characteristic of each LSV for UMAP. Thank you!

Rosa