Use our recommended method, MEME, to look for evidence of both diversifying, and importantly, episodic, selection at individual sites.
Four different codon-based maximum likelihood methods, SLAC, FEL, REL, and FUBAR, can be used estimate the dN/dS (also known as Ka/Ks or ω) ratio at every codon in the alignment. An exhaustive discussion of each approach can be found in the methodology paper.
The codon-based maximum likelihood IFEL method can investigate whether sequences sampled from a population (e.g. viral sequences from different hosts) have been subject to selective pressure at the population level (i.e. along internal branches). A discussion of the method and its application can be found here
All six methods can also take recombination into account. This is done by screening the sequences for recombination breakpoints, identifying non-recombinant regions and allowing each to have its own phylogenetic tree.
Use the RELAX method for detecting relaxed selection in a codon-based phylogenetic framework. Given two subsets of branches in a phylogeny, RELAX can determine whether selective strength was relaxed or intensified in one of these subsets relative to the other.
For coding sequences, the TOGGLE model, developed by Wayne Delport and colleagues, can detect selection-driven changes that result in amino-acid toggling. A canonical example of this can be found in immune-driven evolution of HIV-1 (escape and reversion).
Use the PRIME method to look for site-specific aminoacid properties (e.g. charge, polarity) which are being preserved or modified by the evolutionary process. For example, when a site is positively selected, evolution may be working to change side-chain volume, while maintaining polarity.
Using the modeling framework, which allows the efficient estimations with models which permit dN/dS variation along both sites and lineages, Datamonkey implements a test for finding lineages subject to episodic diversifying selection (EDS).:Branch-site REL method, identifies those branches where a proportion of sites evolves under EDS. If you are primarily interested in finding which lineages (but don't care about which sites) have experienced EDS, use this method.
Deprecated in favor of Branch-site REL. The codon-based genetic algorithm GABranch method can automatically partition all branches of the phylogeny describing non-recombinant data into groups according to dN/dS. Robust multi-model inference is used to collate results from all models examined during the run to provide confidence intervals on dN/dS for each branch and guard against model misspecification and overfitting (method details).
The PARRIS method, developed by Konrad Scheffler and colleagues, extends traditional codon-based likelihood ratio tests to detect if a proportion of sites in the alignment evolve with dN/dS>1. The method takes recombination and synonymous rate variation into account.
The ESD method, described in a 2010 paper, fits a versatile general discrete bivariate model of site-by-site selective force variation to partition all sites into selective classes, and obtains an approximate posterior distribution of this partitioning. The resulting "noisy" distribution of selective regimes is the evolutionary fingerprint of a gene. The EVF (evolutionary fingerprinting) module implements this procedure, and can also infer which individual sites appear to be positively selected while accounting for parameter estimation error (analogous to the BEB methodology of the PAML package).
A Bayesian graphical model is deduced from reconstructed substitutions at each branch/site combination to infer conditional evolutionary dependencies of sites in the alignments, i.e. whether a site is more or less likely to experience a non-synonymous substitution at a branch when certain other sites do (or do not) experience non-synonymous substitutions at the same branch. The SPIDERMONKEY method was introduced in the evolutionary context in our paper on the evolution of the phenotypically important and highly variable V3 loop of the envelope glycoprotein in HIV-1.
Recombination leaves an imprint on sequence alignments: different segments of the alignment may be described by different phylogenetic trees, called phylogenetic discordance. Datamonkey.org implements two methods: SBP, suitable for answering the question "Is there evidence of recombination in the alignment?", and GARD, that attempts to find all the recombination breakpoints. Both method are described in this paper. The output of GARD is accepted by most other analyses, and because recombination can mislead phylogenetic analysis that do not account for it, we strongly urge that recombination testing be done on any alignment that is going to be analyzed for positive selection.
For each type of data, nucleotide, amino-acid and codon, Datamonkey implements separate model selection procedures. An exhaustive search is performed for all possible (Markov, time-reversible) models of nucleotide evolution. For protein data, a collection of published empirical models are fitted to the alignment and the best one is selected using AICc. Finally, for coding data, a sophisticated genetic-algorithm procedure described in our recent paper is used to examine thousands of potential models and report the best one and various metrics based on the set of credible models - this feature is implemented in the CMS module.