Probabilistic Models & Gene Finding

Profile Hidden Markov Models (pHMMs)

Interactive exploration of the Plan 7 architecture, Viterbi optimal path alignment, Forward-Backward posterior decoding, and profile domain searches.

Ready

Speed:

Viterbi Hidden State Trajectory:

Profile Length (K):0

Query Length:0

Alignment Score:0.00

Mode:Viterbi

Plan 7 State Architecture (HMMER3 / Pfam)Hover state node to inspect

2D Dynamic Programming HeatmapHover cell for math breakdown

Plan 7 Architecture

Each column k comprises a Match state (M_k, consensus emission), an Insert state (I_k, background emission with self-loop), and a Delete state (D_k, non-emitting silent state), modeling insertions and deletions with affine-like penalties.

Viterbi vs Forward-Backward

Viterbi uses dynamic programming in O(N · K) time to find the single highest-probability hidden state path. Forward-Backward sums across all possible alignments, calculating exact posterior probabilities for each residue-state assignment.

Baum-Welch & HMMER Searches

pHMMs form the computational backbone of Pfam, InterProScan, and HMMER3. Given unaligned sequence families, Baum-Welch (EM) iteratively re-estimates emission and transition distributions without manual labels.