Naegle Lab Softwarekmn4mj@virginia.edu
ASPEN: Reconstructing protein divergence histories
Understanding the divergence history of proteins has been important to furthering our knowledge of protein function. There are many methods to infer the history of protein duplications, which give rise to paralogs, and their passage onto future species, which gives rise to orthologs.
However, all methods suffer from two fundamental issues: 1) that the true model of evolution is hidden from us, so gauging the accuracy of a model for real protein sequences is impossible and 2) the model inference is extremely sensitive to the selection of input sequences (orthologs in living species) and the alignment method used.
We hypothesized that, despite the variance introduced by using different subsamples of the ortholog sequences in tree reconstruction, consistent relationships across an ensemble of trees would indicate true signal. ASPEN is a method that utilizes this hypothesis to reconstruct trees that are most consistent with observations in an ensemble of trees. We found that reconstructed ASPEN trees are more accurate than the traditional approach of creating one tree from one alignment of all sequences.
Additionally, using a technique to measure the reproducibility of a tree from all sequences (Precision), we find that Precision is a direct proxy for the likely accuracy of the all-sequence tree topology. In conclusion, we find that subsampling from available ortholog sequences is a powerful technique for identifying the likely accuracy of an all-sequence tree, reconstructing more accurate trees, and indicating the number of likely, but diverse, models of evolution one should consider in subsequent analysis or interpretation.
Preprint is available via bioRxiv
Simulation materials are available via Figshare