Quick Lesson on dN/dSWhat does selection “look” like?Neutral SelectionCodon DegeneracyCodon DegeneracySynonymous vs Non-synonymousSynonymous vs Non-synonymousdN/dS ratiosSelection and dN/dSWhy Selection?dN/dS ProblemSLAC single likelihood ancestor countingSLACSLACFEL fixed effects likelihoodFELFELREL random effects likelihoodRELRELSimulation PerformanceSelection and dN/dSQuick Lesson on dN/dSNeutral SelectionCodon DegeneracySynonymous vs. Non-synonymous dN/dS ratiosWhy Selection?The ProblemWhat does selection “look” like?Yokoyama S et al. PNAS 2008;105:13480-13485When moving into new dim-light environments, vertebrate ancestors adjusted their dim-light vision by modifying their rhodopsins•Functional changes have occurred•Biologically significant shifts have occurred multiple times•How do we know whether these shifts are adaptive or random?Neutral SelectionMutations will occur evenly throughout the genome. Pseudogenes?Introns?Promoters?Coding Regions?Codon DegeneracyCodon DegeneracyAA #3AA #2AA #1Wobble effect – an AA coded for by more than one codon1st position = strongly conserved2nd position = conserved3rd position = “wobbly”Pos #3Pos #2Pos #1Synonymous vs Non-synonymousSynonymous: no AA changeNon-synonymous: AA changeSynonymous vs Non-synonymousdN/dS ratiosN = Non-synonymous changeS = Synonymous changedN = rate of Non-synonymous changesdS = rate of Synonymous changesdN / dS = the rate of Non-synonymous changes over the rate of Synonymous changesSelection and dN/dSdN / dS == 1 => neutral selectiondN / dS <= 1 => negative selectiondN / dS >= 1 => positive selectionNo selective pressureSelective pressure to stay the sameSelective pressure to changeWhy Selection?Identify important gene regionsFind drug resistanceLocate thrift genes or mutationsdN/dS ProblemAnalyzes whole gene or large segmentsBut, selection occurs at amino acid levelThis method lacks statistical powerThus the purpose of this paperSLACsingle likelihood ancestor countingThe basic idea:Count the number of synonymous and nonsynonymous changes at each codon over the evolutionary history of the sampleNN [Ds | T, A]NS [Ds | T, A]SLACE40KL10ISLACStrengths:Computationally inexpensiveMore powerful than other counting methods in simulation studiesWeaknesses:We are assuming that the reconstructed states are correctAdding the number of substitutions over all the branches may hide significant eventsSimulation studies shows that SLAC underestimates substitution rateRuntime estimatesLess than a minute for 200-300 sequence datasetsFELfixed effects likelihoodThe basic idea:Use the principles of maximum likelihood to estimate the ratio of nonsynonymous to synonymous rates at each siteFELLikelihood Ratio TestHo: α = βHa: α ≠ βfixedFELStrengths:In simulation studies, substitution rates estimated by FEL closely approximate the actual valuesModels variation in both the synonymous and nonsynonymous substitution ratesEasily parallelized, computational cost grows linearlyWeaknesses:To avoid estimating too many parameters, we fix the tree topology, branch lengths and rate parametersRuntime Estimates:A few hours on a small cluster for several hundred sequencesRELrandom effects likelihoodThe basic idea:Estimate the full likelihood nucleotide substitution model and the synonymous and nonsynonymous rates simultaneously.Compromise: Use discrete categories for the rate distributionsREL1. Posterior Probability2. Ratio of the posterior and prior odds having ω > 1RELStrengths:Estimates synonymous, nonsynonymous and nucleotide rates simultaneouslyMost powerful of the three methods for large numbers sequencesWeaknesses:Performs poorly with small numbers of sequencesComputationally demandingRuntime Estimates:Not mentionedSimulation Performance64 sequences8 sequencesSelection and dN/dSdN / dS == 1 => neutral selectiondN / dS <= 1 => negative selectiondN / dS >= 1 => positive selectionNo selective pressureSelective pressure to stay the sameSelective pressure to
View Full Document