Some sites within a sequence may be more likely to undergo change than are other sites
In other words, the underlying rate of sequence evolution may vary
Also called "site to site rate variation"
For our discussion here, each site is a single character
Invariant Sites Model
Takes into account sites that can never vary
This is sometimes confused with sites that are not observed to vary. In this case the concern is sites that cannot possibly vary.
Assumes characters fall into two rate categories
Not variable
these positions are assumed to never vary under any circumstances
Sites under very strong selection, where any mutation was lethal, would never be observed to vary
Variable
All other sites are assumed to be evolving at the same rate
Realistic, in that there probably are some sites that are effectively invariant, but unrealistic in that the distinction between variable and non-variable is probably not a boolian variable.
Gamma Rate Distribution
Convenient, because it can model a wide range of plausible distributions of rates
Two parameters, alpha and beta
Beta is a scale parameter. If beta = 1/alpha, then mean rate is 1, and alpha is the critical parameter.
If alpha is large (> 200), then all sites evolve at roughly the same rate (i.e., the rates cluster around 1).
If alpha is small (< 1), then many sites evolve slowly, but a few evolve rapidly. This approximates an invariant sites model.
With intermediate values of alpha, there is a broad distribution of rates from slow to fast.
A similar method has been proposed by Van de Peer.
Invariant Sites plus Gamma Distribution
The two models can be combined. Of course the gamma shape parameter alpha will have a substantially different value in a combined model.
Relative Rates
DNArates
Use maximum likelihood to estimate individual rate for each site
LogDet
Models of Amino Acid SubstitutionDayhoff Matrices (Dayhoff, 1979)
Probability of Accepted Mutation (PAM Matrices)
Based on empirical observations of amino acid substitutions
Dayhoff's calculations were based on a very limited number of protein comparisons by today's standards, but more modern calculations of PAM matrices are available
Jones et al., 1992.
Codon Models
Chemical Property Models