Forcing external constraints on tree inference using ASTRAL

Abstract

Background: To account for genome-wide discordance among gene trees, several widely-used methods seek to find a species tree with the minimum distance to input gene trees. To efficiently explore the large space of species trees, some of these methods, including ASTRAL, use dynamic programming (DP). The DP paradigm can restrict the search space, and thus, ASTRAL and similar methods use heuristic methods to define a restricted search space. However, arbitrary constraints provided by the user on the output tree cannot be trivially incorporated into such restrictions. The ability to infer trees that honor user-defined constraints is needed for many phylogenetic analyses, but no solution currently exists for constraining the output of ASTRAL. Results: We introduce methods that enable the ASTRAL dynamic programming to infer constrained trees in an effective and scalable manner. To do so, we adopt a recently developed tree completion algorithm and extend it to allow multifurcating input and output trees. In simulation studies, we show that the approach for honoring constraints is both effective and fast. On real data, we show that constrained searches can help interrogate branches not recovered in the optimal ASTRAL tree to reveal support for alternative hypotheses. Conclusions:The new algorithm is added ASTRAL to all user-provided constraints on the species tree. BackgroundPhylogeny inference requires solving an optimization problem over the space of all trees. The super-exponential growth of the tree topology space makes examining all trees impossible, even for moderately large datasets. As a result, tree inference algorithms have adopted several heuristics strategies, including iterative search (e.g., hill-climbing), used by most maximum parsimony and maximum likelihood methods.An increasingly popular alternative is searching the tree space using dynamic programming (DP). For an

Comprehensive taxon sampling and vetted fossils help clarify the time tree of shorebirds (Aves, Charadriiformes)

“…We hypothesize that the relationships involving the two taxa interacted with other regions of the tree in such a manner that their unconventional positions increased global quartet support despite decreasing it locally. This phenomenon was recently noted by Rabiee and Mirarab (2020), who showed that constrained ASTRAL searches can reveal well-supported clades absent from the main ASTRAL tree at the cost of decreasing the overall quartet score. As a result, we do not consider the relationships of Dromas and Pluvianus to represent genuine instances of gene tree/species tree discordance, but their different resolutions in the species-tree and concatenated analyses may indicate the presence of such conflict in neighboring regions of the tree.…”

Fast and Accurate Species Trees from Weighted Internode Distances

“…Compared to the wASTRAL-h tree it differed in 8 branches, and compared to the TENT and MP-EST* tree it differed in 9 and 8 branches. The ASTRID tree also did not recover the Columbea clade, a clade that has seen strong support in various analyses of this data [12, 52, 36].…”

<strong></strong> Recent Progress on Methods for Estimating and Updating Large Phylogenies

“…The second technique operates by replacing the constraint set that ASTRAL computes from the input with a smaller constraint set. One such approach uses "external constraints", for example partial information about the species tree, in order to reduce the constraint set size (Rabiee and Mirarab, 2020a), an approach we refer to as "ASTRAL-J" to reflect the flag used in ASTRAL for this case. Another approach runs ASTRID on a collection of subsamples of the gene trees, so that each ASTRID analysis of each subsample produces a candidate species tree.…”