Hidden Markov Models for Sequence: Profiles, States and Decoding
A profile hidden Markov model represents a whole family by position-specific emission and transition probabilities rather than by one representative sequence. This stop covers the decoding step that assigns states to a new sequence and what it buys.
What a learner can do afterwards
- Describe the match, insert and delete states of a profile model and what each one emits
- Explain what a single most probable path gives that per-position posteriors do not
- Say why a profile search recovers remote homologues that a pairwise search misses
1 · Read
A profile hidden Markov model represents a whole family by position-specific emission and transition probabilities rather than by one member sequence. It trains on a multiple alignment and learns which positions are conserved and where gaps tend to occur. Conserved columns reward the expected residue strongly, while variable columns accept almost anything. You already know substitution matrices price single pairs; this model prices every column on its own terms.
Suppose a new sequence arrives with extra residues between two consensus columns and one column missing. The model walks it through states: match states stand for consensus columns and emit the aligned residue, insert states absorb the extra residues between columns, and delete states silently skip the missing column. That walk is an alignment plus annotation in one pass.
Decoding means choosing a path through these states, and two answers serve different needs. The Viterbi algorithm returns the single most probable full path: one coherent global story. Posterior decoding instead reports, for each position, the probability of each state summed over all possible paths: per-position confidence. One commits, the other hedges with numbers.
To argue for a remote homologue that pairwise search misses, point at the adding up. Weak but consistent signals across many positions accumulate under position-specific scoring until the family verdict turns confident. Say which decoding you used and why: Viterbi for one annotation to act on, posteriors where each position's confidence matters.
Profiles score each column separately, Viterbi commits to one path, and posteriors grade every position.
2 · Watch
Take it off screen
Where it sits
Learn first
This opens up
Nothing builds on it yet.
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.