Evidence That Inconsistent Gene Prediction Can Mislead Analysis of Dinoflagellate Genomes

Abstract

Comparative algal genomics often relies on predicted genes from de novo assembled genomes. However, the artifacts introduced by different gene-prediction approaches, and their impact on comparative genomic analysis remain poorly understood. Here, using available genome data from six dinoflagellate species in the Symbiodiniaceae, we identified methodological biases in the published genes that were predicted using different approaches and putative contaminant sequences in the published genome assemblies. We developed and applied a comprehensive customized workflow to predict genes from these genomes. The observed variation among predicted genes resulting from our workflow agreed with current understanding of phylogenetic relationships among these taxa, whereas the variation among the previously published genes was largely biased by the distinct approaches used in each instance. Importantly, these biases affect the inference of homologous gene families and synteny among genomes, thus impacting biological interpretation of these data. Our results demonstrate that a consistent gene-prediction approach is critical for comparative analysis of dinoflagellate genomes.

Genomes of the dinoflagellate Polarella glacialis encode tandemly repeated single-exon genes with adaptive functions

“…As expected, the number of predicted genes in all six (haploid) genomes of Symbiodiniaceae is roughly similar to the rough estimate of haploid gene number for the two P. glacialis genomes ( Table 2 and Additional file 3: Supplementary Table 12). The proportion of genes predicted in P. glacialis that are supported by transcriptome evidence (~94% for each isolate; Table 2) is much higher than in the Symbiodiniaceae isolates (~79% averaged among six genomes [39]). This result may be explained by the more extensive transcriptome data we generated in this study (using both RNA-Seq short-read and Iso-Seq full-length transcripts) to guide our gene prediction workflow (see the 'Methods' section), compared to the transcriptome data (based on RNA-Seq short-reads) available for the other isolates.…”

“…Prediction of protein-coding genes in Polarella glacialis is likely impacted by RNA editing Using a gene-prediction workflow customised for dinoflagellate genomes [39] (see the 'Methods' section), we predicted 58,232 and 51,713 protein-coding genes (hereinafter genes) in the CCMP1383 and CCMP2088 genomes, respectively ( Table 2 and Additional file 3: Supplementary Table 12). Of the 58,232 genes predicted in CCMP1383, 51,640 (88.68%) of the encoded proteins were recovered in CCMP2088 ( Fig.…”

“…50,000 genes in a diploid assembly versus 25,000 in a haploid assembly); this number remains a rough estimate, because the delineation of a haploid P. glacialis genome remains to be systematically investigated with genome data at chromosomal resolution. The workflow used to predict genes in the Symbiodiniaceae genomes is the same as used in this study [39], making the predicted gene features more comparable. As expected, the number of predicted genes in all six (haploid) genomes of Symbiodiniaceae is roughly similar to the rough estimate of haploid gene number for the two P. glacialis genomes ( Table 2 and Additional file 3: Supplementary Table 12).…”

“…Scaffolds were retained and considered non-contaminant, if ≤ 10% of their length was covered by BLAST hits with a bit score > 1000 and E ≤ 10 −20 . Further identification and removal of contaminant sequences were conducted following Chen et al [39]. Scaffolds were removed if they had a G+C content and length outside the expected normal distribution; no putative contaminant scaffolds were identified using this approach.…”

“…To predict protein-coding genes, an ab initio gene prediction approach adapted from Liu et al [12] was applied to the genomes of P. glacialis; this approach, also reported in Chen et al [39], is available at https://github. com/TimothyStephens/Dinoflagellate_Annotation_ Workflow.…”