The shape size for the kmer_hash.

[Jappy0]

Hi,

Many thanks for this excellent library in computational biology. I have a question when I was using the API. Please see my question below.

Platform

• SeqAn version: 3.3.0
• Operating system:
• Compiler: gcc13.20

Question

I wanted to hash each kmer of a sequence using the hash method in Seqan3. The used kmer size is 28, so the shape is also 28. but

"terminate called after throwing an instance of 'std::invalid_argument' what(): The chosen shape/alphabet combination is not valid. The alphabet or shape size must be reduced."

However, I read the documentation; it says the maximum size of the shape is 58.

So, is all my understanding correct? or is this a bug?

#include <seqan3/search/views/kmer_hash.hpp>

std::vector<seqan3::dna5> read{"CGCTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTGCGTCGTCATCGCTGTCATCCCACATGCCCAGCAGTGCTGATCTGGGACAGCTTGGCGTAACTAGA"_dna5};
unit8_t k=28;
for (size_t i = 0; i < read.size() - k + 1; ++i){
        auto kmer = std::vector<seqan3::dna5>(read.begin() + i, read.begin() + i + k);
        auto hash_val_range = kmer | seqan3::views::kmer_hash(seqan3::ungapped{static_cast<uint8_t>(k)});
}

Thanks for your time.

Best regards,

Jappy

[SGSSGene]

Hi, thank you for reaching out.

As far as I see there are two sources:

1. Probably the one you are referring to Shape
2. An one inside the kmer_hash Which warns against the exact exception you get.

@eseiler and @smehringer This seems to be much more your field of expertise than mine. What do you think about this?

Complete working example, including the exception being thrown:

```c++ #include #include using namespace seqan3::literals; int main() { std::vector read{"CGCTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTGCGTCGTCATCGCTGTCATCCCACATGCCCAGCAGTGCTGATCTGGGACAGCTTGGCGTAACTAGA"_dna5}; uint8_t k=28; for (size_t i = 0; i < read.size() - k + 1; ++i){ auto kmer = std::vector(read.begin() + i, read.begin() + i + k); auto hash_val_range = kmer | seqan3::views::kmer_hash(seqan3::ungapped{static_cast(k)}); } } ```

[SGSSGene]

Independent of the documentation.

A k-mer over dna5 will require at least log_2(5) ≈ 2.3219 bits, which means shapes with max 27 ones are possible. If you changed it to dna4, you could go up to 32 ones inside your shape.

[Jappy0]

Independent of the documentation.

A k-mer over dna5 will require at least log_2(5) ≈ 2.3219 bits, which means shapes with max 27 ones are possible. If you changed it to dna4, you could go up to 32 ones inside your shape.

Thanks so much for the explanation. I now have a better understanding.