Dataset: Genome assembly: Montastrea cavernosa

Our assembly approach follows a similar method used to generate a reference genome for the coral Acropora millepora (Fuller et al., 2020). We first constructed an initial draft assembly using only PacBio long reads with canu v1.6 (Koren et al., 2017). We then performed two rounds of scaffolding on this PacBio-only assembly using the 10X paired-end reads with scaff10x v1.1 (https://github.com/wtsi-hpag/Scaff10X). The resulting assembly underwent additional scaffolding using the 10X barcodes and paired-end reads with the ARCS/LINKS pipeline (Warren et al., 2015; Yeo et al., 2018). The PacBio long reads were then mapped to the assembly and used to join contigs with SSPACE-longread v1.1 (Boetzer & Pirovano, 2014). A final two rounds of scaffolding were then performed using scaff10x v1.1 again. We filled in gaps with PBJelly v15.8.24 (English et al., 2012) and error corrected with PacBio reads using the Arrow v5.0.1.9585 algorithm (https://github.com/PacificBiosciences/GenomicConsensus). To create the final assembly, we used the paired-end reads with 10X barcodes removed to polish with pilon v1.22 (Walker et al., 2014). In total, this resulted in 10,835 scaffolds with an N50 of 248Kb and maximum length of 1.8Mb.

An initial all-vs-all alignment of the assembly revealed the presence of redundant contigs. To further refine the assembly, we used the HaploMerger2 (Huang et al., 2017) pipeline to rebuild both haploid sub-assemblies from the potential polymorphic diploid genome assembly. Afterwards, the final phased 448Mb assembly contained 5,161 sequences, with an N50 of 343Kb and average length of 86kb. The maximum scaffold length was unchanged.

Protein coding gene models of the M. cavernosa genome were annotated using MAKER-P v2.31.9 (Campbell et al., 2014, Cantarel et al. 2008; Holt & Yandell 2011). MAKER-P was implemented using the pipeline available at the CyVerse website (http://www.cyverse.org/), available as an Atmosphere image and MPI-enabled for parallel processing. To predict gene models, we used previously published transcriptome data (SRA accession: SRP063463) (Kitchen et al., 2015) and mapped these reads using a reference-based assembly strategy with the TopHat v.2.1.1 and Cufflinks v2.2.1 pipeline (Trapnell et al., 2012), resulting in 43,577 transcripts. In addition, we generated a de novo transcriptome from the raw RNA-seq data with Trinity v2.0.2 (Grabherr et al., 2011), producing 200,233 transcripts. Both sets of transcripts were used as evidence in the MAKER-P pipeline.

For protein-level evidence in the annotation pipeline, we used peptide sequences from 8 robust coral species: Pseudodiploria strigosa, Seriatopora hystrix, Platygyra carnosus, Montastraea faveolata, Madracis auretenra, Fungia scutaria, Favia sp, and Montastraea cavernosa itself (Bhattacharya et al., 2016). We excluded other relatively evolutionary distant coral species due to the significant divergence between genomes, as they are likely to provide only limited evidence for gene annotation.

Before running MAKER-P, we also performed de novo repetitive element identification for the genome assembly using RepeatScout v1.0.5 (Price et al., 2005). The assembly was then soft masked with this resulting repeat library using RepeatMasker (http://www.repeatmasker.org, RepeatMasker 4.0.7 and Library, Smit et al., 2013-2015). 44.7% of the genome was masked and identified as repetitive. To generate the final gene set, we performed three rounds of training and predicting gene models with MAKER-P, in combination with AUGUSTUS (Hoff & Stanke, 2019). In total, 30,390 non-redundant gene models were produced from the pipeline. We assessed the completeness and quality of the predicted final gene set using 978 near-universal single-copy orthologs from the Metazoan set available in BUSCO v3 (Simão et al., 2015; Waterhouse et al., 2018). Of these, 793 are completely present and 96 are present as fragments, for a total of 90% matches to BUSCO groups searched. Finally, functional annotation was performed for the gene set using the Trinotate pipeline (Bryant et al., 2017) and eggNOG-mapper (Huerta-Cepas et al., 2017).