How SPAdes prevents self-loops while breaking sequences into even-sized words?

[apcamargo]

In a previous discussion, it was explained that in SPAdes, the overlaps in the assembly graph have a length of k, and sequences are broken down into (k+1)-mers. In my assemblies, the overlap length is indeed 55 bp.

However, my understanding is that, for assembly, sequences should be broken into odd-sized k-mers to prevent self-loops in the graph, caused by palindromes. This would result in even-sized overlaps. For example, in Minia, the overlaps are (k-1) bp long, where k is an odd number.

Given that SPAdes uses (k+1)-mers, how are self-loops avoided in the assembly process?

[asl]

There are two different "flavours" of de Bruijn graph, so-called "edge-centric" and "node-centric". SPAdes employs edge-centric approach. See for example https://www.biostars.org/p/175058/ for some explanation

[apcamargo]

Thank you!