In order to analyze the coverage of the subsequences over a sequence, it is helpful to know their positions. There are two ways to do so:
a) Write a second function to every function returning the positions instead of the subsequences.
b) Write one function, getting the sequence and the subsequences and align them in order to get the positions.

The drawback of b) is that it takes more time to compute because the initial function needs to be called and then an alignment is applied, but its advantage is that it is only one function to implement, which then should work with every method.

For now, I think b) is the easier option. Should it be necessary to get the positions in a more efficient way, a) can still be implemented.

It seems that including the reverse strand by selecting a minimum value of hashes is not doable since the syncmers are being chosen based on the Smers. A syncmer is a Kmer that has its smallest smer at its start or end and in a sense, Smers should correspond to Kmers to make this work. Hence, it is not possible to select minimum hashes of Smers and Kmers independently. It is possible to select the minimum kmer_hash first and then check its corresponding Smers but this would be time-consuming and needs double the number of vectors, ranges, and iterators.

Add Raptor as a submodule to make the same comparison as in the Raptor paper.
Note: The search probably needs to be adapted to the different methods.

The syncmer implementation is not easily compatible with gapped k-mers, because we would need the actual shape used. Unless the views of smers could be adapted beforehand accordingly.

It is not quite clear, what to do, if the minimum appears multiple times in a submer, what is then the minimum? The first occurrence or all occurrences?
If I understand GetMinSubkmerPos in https://github.com/rcedgar/syncmer/blob/master/kmer.cpp correctly, then only the first occurrence is seen as minimum. Therefore, for an offset of 1 and k=4, s=2 the sequence AAAA would not be a syncmer.

• syncmers
• strobemers (has its own library, can just be added by that?)
• weighted minimizers

Finding a good and representative set of k-mers

• universal hashing sets
• polar sets
• fleximers
• sig-mers
• deepMinimizer

Error relates to debug_stream and that it can not print the range, which should be capable of.

The methods are compared by their capability to represent the sequence data as well as their performance.

• distribution
• minimal, average and maximal gap between elements
• speed
• RAM usage
• number of elements created (compression factor? number of elements divided by number of k-mers)
• capability to find transcripts (True Positives, False Positives, True Negatives, False Negatives)
• conservation (how many minimizers stay the same when sequence is slightly mutated, conservation should not be based solely on numbers of minimizers, but counting number of nucleotides covered in order to account for overlapping minimizers, see syncmer paper)
• distance on mutated sequence compated to not mutated sequence
• Handling of repitive and errorneous k-mers (simple cutoffs vs none vs weighting)

In the strobemer paper, they use two simulated data sets and define:

• the number of matches, where a match between sequence A and mutated sequence A' is an idential subsequence at position i

• the positions covered by the subsequences

• an island: a maximal interval of consequtive positions not covered as an island

• analysis how many submers are unqiue in a given genome and how similar these submers are to each other (edit distance)

Also, they check, how many unique subsequences there are in the five largest human chromosome to measure the precision of a method.