Apply mask before argmax

[aerval]

Dear Kishore,

Thanks for the update. Can you explain the masking feature a bit more. The way I understand the code is that you choose for each task the position with the highest predicted effect and then check whether that position is at a known splice junction. This goes against my intuition since I would mask the position of the splice junction (or everything else for a junction) and then look for the highest effected other (unmasked) position. Further, it seems to me that you are only checking the closest splice junction which, especially for small exons, is not necessarily the only one to consider?

Cheers, Philipp

[kishorejaganathan]

Your intuition of the masking feature is correct. I agree with you on both masking before looking for the highest effect position and checking more than just the closest splice junction. In the code which was used to annotate all SNVs, I ended up implementing the mask this way as it was a quick hack which worked most of the time and now I'm unfortunately stuck with it (at least in this release) as I wanted one version of the github tool which could precisely reproduce all the pre-annotated files. In the next release, I intend to re-implement the mask precisely the way you mentioned. Thank you for bringing this up nevertheless.