merge k-mers of different lengths

[Leo-ccc]

Hi,

The lengths of the features (or k-mers) I'm interested in are different. It didn't return any warning in the 'index' processing, but I got an error message when I ran the 'merge' module.
[ERROR] KaMRaT-merge relies on the index in k-mer mode, please rerun KaMRaT-index with -klen option

I wonder if it will be possible to update a method in the future that can merge k-mers of different lengths to obtain contigs?

[hl-xue]

Hi,

Thanks for using Kamrat.

Currently, Kamrat does not support merging features with variable length, as our definition of k-mer is “sequence of fixed length k”. Also, when kamrat index is launched without the argument -klen INT, the features are treated as general character strings (i.e., not absolutely being sequences A/C/G/T but can be gene names such as TP53, transcript IDs such as ENST00000714409, etc.). So, it doesn’t check the feature length and thus doesn’t warn k-mers having variable lengths. However, normally it should raise a warning "indexing in general: features are not considered as k-mers” in the index step. Please let me know if this is not the case.

I imagine a potential solution to your demand: maybe firstly launching a k-mer counting method (e.g., jellyfish) to break the variable-length sequences into fixed-length k-mers, then joining the count vectors into a count table would help. After this, you should be able to run Kamrat on the newly obtained k-mer count table (please run with kamrat index -klen INT to define the analysis in k-mer mode).
As for k-mer length, I would suggest targeting for k=31 if possible, or alternatively an odd number not shorter than 21.

Potentially in a future version of Kamrat, we may support the merge function with variable k-mer length. This is within our long-term plan.

Please feel free to comment below if any further questions.

Kind regards,
Haoliang