Interactive demo · Statistical Methods for Modern Biology · Chapter 10

Substitution models and evolutionary distances

Along a branch, the base at a site changes at random times as a continuous-time Markov chain with rate matrix Q; after d expected substitutions per site the transition probabilities are P = exp(Qt) (left). Two sequences that diverged by d differ at a fraction p of their sites, and p grows more slowly than d because later substitutions hit sites that already differ, or revert them (middle): the Hamming proportion underestimates the distance, more and more as d grows (Holmes and Huber 2019, Question 10.10). Inverting the curve gives the Jukes-Cantor (JC69) or Kimura (K80) distance. The bottom panel applies the three distances to 20 aligned HIV gag sequences (Exercise 10.8).

Model

κ = α/β

True distance d (substitutions per site)

Sequence length L

Simulate pairs of sequences

Scoreboard

Try these

JC69: every substitution has rate α, d = 3αt. K80: transitions (A↔G, C↔T) have rate α, each transversion rate β, d = (α + 2β)t. A simulated pair has L independent sites; each differs as a transition with probability P(d) and as a transversion with probability Q(d). The estimate is undefined (infinite) when the observed proportions fall beyond the saturation level. The gag distances use pairwise deletion: a site counts for a pair only if both bases are A, C, G or T.

In the lecture notes