kmerseek docs

Interactive pages on how kmerseek scores a region and judges whether it could be chance. Every slider recomputes the equations and figures.

Part 1: How a kmerseek region gets an E-valueThe region score, the chance match rate, nats per score point, the E-value, and how K and r_database are measured when the index is built. Part 2: What two letters can carryThe copy rate, bits per position, the trade between alphabet size and region length, short motifs, summing several regions, and the best mismatch penalty. Part 3: What every alphabet can seeCed9, P66 and BHF searched against the human proteome with all 19 alphabets: where each alphabet ranks the known partner under five metrics, the seed information an E-value of 1 needs, and why many regions get no E-value.

What a search looks like

C. elegans CED-9 searched against 25 human BCL-2-like proteins, hydrophobic/polar alphabet, k=12. The same pages scripts/visualize_search.py and scripts/visualize_pair.py write for any search.

CED-9 against 25 BCL-2-like proteinsOne row per protein, every run drawn on the CED-9 axis. Click a row for its dot plot, its alignments and both full sequences. Sort, filter, and search for a sequence such as NWGR. CED-9 against BCL-227 shared 12-mers. The longest run is the BH1 motif: 19 residues, 5 identical, inside the Bcl-2 domain of both proteins. It lies on the residue pairs USalign gets from the two AlphaFold models (TM-score 0.48). CED-9 against BCL-xLThe same BH1 run in another family member: 19 residues, 6 identical, TM-score 0.47. Both homologs put their longest run over the same stretch of CED-9. CED-9 against reticulon-3100 shared 12-mers against BCL-2's 27, spread over 22 runs of a 1032-residue protein. The longest, 23 residues, has 2 identical residues and sits 40 residues off the USalign pairs. Sharing more k-mers than a homolog does not make a match.