Hidden Markov Models for Sequence: Profiles, States and Decoding · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

One model for a whole protein family

Science · Genetics & Evolution · ages 22-23
Name ______________________   Date ____________

  1. A sequence skips a consensus column entirely. Which state handles this?

    • The delete state
    • The match state
    • The insert state

  2. What does a profile model store about a protein family?

    • One representative sequence only
    • Position-specific emission and transition probabilities
    • The names of all biologists

  3. A delete state emits extra residues between consensus columns.

    Circle one:   True   False

  4. Two positions both land on state X by Viterbi, but one has 99 percent posterior and the other 51 percent. What does this tell you?

    • Both positions are equally certain
    • The 99 percent position is far more trustworthy than the 51 percent one
    • Posteriors are meaningless numbers

  5. You need one complete annotation to act on, not per-position doubts. Which decoding do you run?

    • Posterior decoding
    • No decoding at all
    • Viterbi decoding

  6. Pairwise search misses a match that profile search catches. What best explains the rescue?

    • Profiles add up weak consistent signals across many positions
    • Profiles use shorter sequences
    • Profiles ignore conservation entirely

  7. A highly variable column should strongly reject unexpected residues, just like a conserved column does.

    Circle one:   True   False

  8. A student reports only the Viterbi path and claims every position is certain. What is the error?

    • Viterbi paths are always wrong
    • Certainty needs no numbers
    • One best path says nothing about per-position doubt; posteriors are needed for that
LightMySky · lightmysky.comW1-mt_nk14IaZTcw-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

One model for a whole protein family W1-mt_nk14IaZTcw-s1

  1. The delete state · Delete states silently skip columns a sequence lacks.
  2. Position-specific emission and transition probabilities · Profiles learn each column's conservation and gap habits.
  3. False · Deletes emit nothing; inserts emit the extras.
  4. The 99 percent position is far more trustworthy than the 51 percent one · One path hides doubt that per-position confidence reveals.
  5. Viterbi decoding · Viterbi commits to the single most probable full path.
  6. Profiles add up weak consistent signals across many positions · Position-specific scoring accumulates faint agreement into confidence.
  7. False · Variable columns accept almost anything; only conserved ones reward strictly.
  8. One best path says nothing about per-position doubt; posteriors are needed for that · Commitment is not confidence; the path hides close runner-up paths.
Worksheet · LightMySky