Generate kmer profiles from a bunch of peptide sequences

Question

[Owen S.]

Can anyone recommend a software solution to do this:

• Input: about 100,000 short peptide sequences -- unaligned -- of varying lengths, but mostly under 20 residues.
• Output: amino-acid profiles (e.g. sequence logo map) describing similar over-represented kmers (say, 3-or 4- or 5-mers).

I can think of ways to tackle this myself*, but why re-invent the wheel? Hoping that my question and any discussion that follows may also help others.

Thanks!

PS. My approach would be something like this:

1. count all unique kmers
2. calculate pairwise distances
3. select clusters (clades) of similar kmers
4. use these kmers (and their counts) to build sequence logo maps

Answer 1

[Sean Davis]

The Biostrings Bioconductor package has fast kmer counting (oligonucleotideFrequency) functionality. You can then take your results and do all kinds of stats, clustering, and visualization.

[Owen S.]

Thanks, but my question relates to peptide, not nucleotide, sequences. (The Biostrings function you suggested only works with nucleotide seqs.)