A metaanalysis of bat phylogenetics and positive selection based on genomes and transcriptomes from 18 species

Hawkins, John A.; Kaczmarek, Maria E.; Müller, Marcel A.; Drosten, Christian; Press, William H.; Sawyer, Sara L.

where some subset of species has unanimous agreement of the amino acid sequence but disagrees with the majority of species. This is correlated with but distinct from the alternate-consensus runs defined for the MIXR algorithm as it is independent of the substitution model used for the MIXR scoring function. The results after each step of the cleaning process are shown in Fig. 2D . We see that the most significant reduction in secondconsensus runs is due to the exon structure filtering step, and MIXR cleaned up essentially all of the runs which passed the previous two steps. After cleaning, only seven runs of more than three columns were detected, and those were up to only five columns in length. Furthermore, all but one of these alignments contained only species with genome data available, removing the concern of separation by data type. As a second measure of alignment improvement, we checked that the filtered sequences had increased overall sequence conservation. Our strategy, as expected, preferentially discriminates against more weakly conserved sites [as defined by CLUSTAL (47) ], filtering >80% of nonconserved sites vs. 44% for unanimous sites, and almost all gapped sites (Fig. 2E) . Furthermore, V T V P S S SA A GT L F R G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L F R G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L F R G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GG SR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GG SR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GG SR DP S V T V P S S SA A GT L F R G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GGGR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GG SR DP S V T V P S S SA A GT L FQ G L C GA P DA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S -SC P Q L Q C C R H L V P GP L WC SDA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S -N C P Q L Q C C R H I V P GP L WC SDA P H P L SK I P GGR GGGR DP S Multiple sequence alignment cleaning. (A) Some multiple sequence alignments were observed to demonstrate isoform selection biased toward separation of genomic and transcriptomic data, causing nonrandom, nongapped errors segregated by the artifact of data type. (B) The alignment cleaning pipeline. First, each gene derived from genomic data was revisited to choose the isoform which best matches the consensus alignment sequence. Second, exons were filtered by structure, where exons with disagreement about boundary positions in the alignment and exons with >1% length difference between species were filtered out. Last, exons were filtered using the MIXR algorithm described in the text. (C) A total of 3,444 improved genomic isoform selections were performed, shown as a function of the improvement in percent matching the consensus sequence, i.e., (percent matching after) -(percent matching before). (D) Counts of unanimous second-consensus runs after each step of the alig