AnnotationBRAKER

This repository consists of a Snakemake workflow to produce an annotation of protein-coding genes in a genome with BRAKER. BRAKER3 is the latest code release and the paper describing how it works can be found here. BRAKER is a method for genome annotation that uses extrinsic evidence to guide optimized model parameterization andcoding sequence feature prediction on a genome sequence with Hidden Markov Model (HMM)-based tools, primarily AUGUSTUS,and, to a lesser extent, GeneMark. While BRAKER has a large number of command line options for customizing genome annotation, there are three main ways to predict genes with BRAKER:

• Only using RNA-seq evidence, previously known as BRAKER1
• Only using protein evidence, previously known as BRAKER2
• Using both RNA-seq and protein evidence, aka BRAKER3. BRAKER3 runs separate protein-only and RNA-seq only gene prediction analyses, then merges them by selecting transcripts using a scoring schema with TSEBRA.

We have found that BRAKER3 frequently outperforms BRAKER1 (see our preprint for results supporting this claim). Thus, when RNA-seq data are provided, this workflow runs BRAKER3, i.e there is currently no option to run BRAKER1.

Software requirements

To get this workflow to work, uou need to install Snakemake. We recommend doing this into a conda environment, using a MiniForge python distribution:

mamba create -n snakemake python=3.11
mamba activate snakemake
mamba install snakemake
mamba install bioconda::snakemake-executor-plugin-slurm
mamba install conda-forge::singularity 
mamba deactivate

When running this workflow, it will have to be done from inside the snakemake conda environment so that Snakemake is accessible. This workflow was developed to run on a cluster that uses SLURM as the job scheduler, and we have provided a SLURM job-runner script braker_snakemake_slurm_runner.sh , that can be modified for use with other schedulers. In addition, this workflow runs BRAKER from inside a singularity container which is installed on the fly. While we provide an example slurm profile config.yaml file, if you choose to modify this, or are running on a different HPC scheduler, you will need to make sure to specify the same singularity-related variable values in the profile configuration file.

Protein evidence

Protein evidence for BRAKER2 and BRAKER3 should be in the form of a fasta file, derived from OrthoDB, using a taxonomically relevant subset of the complete database. A convenient way to generate such fasta files is using orthodb-clades, written by the lead developer of BRAKER2. While this Snakemake workflow by default generates a handful of clade-level fasta files, as well as several species-level ones, it is straightforward to edit the underlying code to produce outputs customized to your taxonomic needs. In our experience, this is far easier than attempting to use the OrthoDB API.

RNA-seq evidence

This workflow requires paired-end RNA-seq reads, and these can be supplied as fastq files or as spliced alignments to the genome in bam format. When fastq files are supplied, this workflow aligns the reads with STAR, and Dobin et al., 2013 in 2-pass mode, and this includes the construction of the STAR mapping index.

At present, this workflow does not allow for the use of long RNA-seq reads such as those generated with Pacific Biosciences or Oxford Nanopore Technologies sequencing instruments. With short reads, we have found that direct assembly of reads into transcripts often leads to higher quality genome annotations than with BRAKER. See our preprint that provides results supporting this claim. We suspect the performance advantage for direct assembly will be even greater with long RNA-seq reads, although this needs to be confirmed with additional testing. We strongly suggest that, when you have paired-end RNA-seq data, that you also generate annotations with Stringtie combined with TransDecoder, and compare your outputs from BRAKER with this approach. It can be easily implemented in another of our Snakemake workflows found here.