Sequence Alignment, Parameter Sensitivity, and the Phylogenetic Analysis of Molecular Data

Abstract

The dependence of the results of molecular phylogenetic sequence analysis (both align­ ment and cladogram construction) on variation in analytical parameters is examined. Phylogenetic analyses of molecular sequence data are necessarily based on intrinsically unmeasurable param­ eters such as transition-transversion and alignment gap cost ratios (among others). Procedures for robust and liberal hypothesis choice are proposed using congruence as an optimality criterion. To illustrate and explain this process further, data on arthropod relationships are used. The effects of variation in transversion-transition and gap-change ratio parameters on alignment and phy­ logeny reconstruction are assessed in light of both taxonomic and character-based congruence measures. (Sequence alignment; sensitivity analysis; arthropods; molecular systematics; phylog­ eny.) The phylogenetic analysis of nucleic acid sequences, as with other data, is unavoid­ ably based on explicit and implicit as­ sumptions. At the fore are character trans­ formation models-usually transversion­ transition ratios-and the relative cost of alignment-derived sequence gaps. These values are the fulcra of sequence analysis. Simple homogeneous weighting does not avoid the issue of arbitrary, yet crucial, as­ sumptions. Transversion-transition ratios and alignment gap costs are generally not directly measurable. These values are statements of process, and they can only be inferred appropriately from a predeter­ mined phylogenetic pattern. The disturb­ ing circularity of the interaction between the specification of values a priori and their inference a posteriori is a general and cen­ tral problem in molecular phylogenetic analysis. One potential solution to the problem of parameter sensitivity has been proposed by Farris (1969; amplified by Carpenter, 1988) through the successive approxima­ tions weighting (SAW) procedure. Iteration is used to estimate parameters repeatedly (in this case character weights) by recon­ structing phylogeny and using this phy­ logeny to generate new self-consistent pa­ rameter estimates. This process is reprised until stability in inferred weights is achieved. The SAW approach has been ex­ tended for character transformation weights by Williams and Fitch (1989). Al­ though an iterative approach is in some sense objective, it will not yield informa­ tion as to how sensitive the results are to the specific model (set of analysis param­ eters) the process yields. Iteration is a way to choose some models over others, but it does not tell us how much better these models are. Furthermore, all iterative ap­ proaches are to some degree sensitive to the initial conditions (a priori weights) of the analysis. Even though transversion-transition and gap-change cost ratios are unmeasur­ able in the absence of· a predetermined phylogeny, it is possible to estimate their values through appeal to an external op­ timality criterion. The most reasonable op­ timality criterion for phylogenetic analysis must be congruence (whether taxonomic (Nelson, 1979) or character based (Mick­ evich and Farris, 1981); but see Miyamoto, 1981, 1985). Without any way of objectively measuring the accuracy of reconstruction, only precision (the agreement among data) can be used to arbitrate among competing hypotheses. This same sort of precision (in the guise of congruence) can be used to assay both the quality and robustness of phylogenetic hypotheses.