Update ref_majority_dels with minimum count for majority base

[SarahNadeau]

As per the discussion on slack we'd like to have a minimum base count for the majority base. I think we just need a line here that insists the argmax of counts is at least min_count = 5.

[sposadac]

Thanks a lot for the suggestion. By default, when the read depth is less than 50 reads, we report the base using lower case. This is not doing exactly what you've asked, but there is another parameter:n_threshold. By default, this parameter is not used, but it can be configured using the cli and it will solved this issue:

smallgenomeutilities/scripts/extract_consensus

Line 136 in 1283457

mask_n = read_count < n_threshold

[SarahNadeau]

As discussed on slack, the clarification of the problem was:
We want to turn a position with reads {A, A, A, T, T, T, T} into "N" because although coverage is 7 >= 5, the majority base T has only 4 < 5 reads supporting it. n_threshold unfortunately counts reads per position, not per base.

Then the issue is where is best to implement this fix (lowercase letters in function majority_vote for positions failing the new criterion + a postprocessing mask, or mask in function majority_vote).

I see the point that it's messy to change the logic of majority_vote when there's already a postprocessing function to do masking. However, I would find it confusing if lowercase letters mean <50x coverage and/or <5x support for the majority vote.

What about instead of (or in addition to) value majority_idx, having majority_vote return majority_count (the # reads supporting the majority base). This could then be accepted as an argument to mask_low_quality, which would handle the masking if majority_count is less than min_coverage. Would that make sense?

[sposadac]

I think of it in a more general sense, what's the advantage of using the read counts of individual bases, as opposed to the read depth per position. Let's consider a bad scenario were the total read depth at a given position is 50 reads, a base would need at least 10 reads (for an alphabet with 5 characters) to be returned as the consensus base using the majority-vote rule. Then, requiring at least 5 reads means that the read depth at that position was between 5 and 25 reads, which seems rather low to me.
My general reasoning is that regions with few read counts should be regarded as low-quality regions and how (e.g., total read count per position vs. base counts) to set the threshold is anyways a bit of a preference/choice.