Ambiguous reads unmapped disappear with FMD Salmon?

Question

[Sharon]

Hi All

If we have too many ambiguous reads unmapped but their length is not short, could this be a reason of reads overlapping too long in the middle, so there is no enough bases at both sides of the read to differentiate the mapping?

And also why those disappear by using FMD salmon mapping the default values but never disappear with Quasi mapping different k indices?

I am just trying to understand more and make sure the high mapping is genuine.

Thanks

[GenoMax]

Now we know the practical implication of @Rob's comment ( C: Alignment dropped with Salmon ). Initial seed length must make the difference in whether a read gets mapped.

[GenoMax]

Tagging: Rob

[Sharon]

Does this mean (from Rob's comment on the post you mentioned) that FMD increases the default mismatches allowed? That'w why we have higher mapping? What do you mean by seed's length.

[GenoMax]

Minimum size of exact match is smaller for FMD so I assume that increases chances of a read mapping with those indexes. @Rob can comment on the role mismatches play in salmon mapping.

BTW: Did you ever look at the resulting counts with both indexes for a specific sample? Were they similar or very different?

[Sharon]

The counts increases with FMD but relatively. Like the ratio between control and tumor remains the same. That's by eye notice. But there is a huge difference between no. of expressed genes if I annotate using NCBI than Genocode reference. NCBI are much more.