Braker Genome Annotation

Braker Genome Annotation is a fully automated pipeline that uses RNA sequencing data to train ab initio gene predictors and generate high quality protein coding gene annotations for eukaryotic genomes. This guide is written for bioinformaticians, genomicists, and graduate students who have a draft genome assembly and RNA sequencing reads and want a reproducible, evidence guided annotation without manually curating training sets. We cover core concepts, decision points, a practical workflow, quality checks, common mistakes, and the limits of interpretation. For a comprehensive route map through BRAKER and related tools see the protocol from Methods in Molecular Biology.

EMBL EBI Training provides foundational resources on eukaryotic genome annotation that complement the material here.

At a Glance

Decision Criteria: When to Use BRAKER

BRAKER is the tool of choice when you have a reasonably complete genome assembly and matching RNA Seq data from the same species or a close relative. It is not a one size fits all solution. Decide based on these factors.

Assembly quality matters. If your assembly has high fragmentation (N50 < 10 kb) or many misassemblies, BRAKER will produce many false positive fragments. Consider assembly improvement first (NCBI Bookshelf for assembly quality guidelines).

RNA Seq representativeness is critical. BRAKER uses spliced alignments to train gene models. If you have only one tissue or single developmental stage, you will miss genes expressed elsewhere. Pool multiple libraries if possible.

You need a repeat softmasked genome. BRAKER works best when repetitive regions are masked (using RepeatModeler and RepeatMasker). Without masking, transposable element fragments are predicted as false genes.

Alternative pipelines exist. For species with very large genomes without RNA Seq, consider Galba or TSEBRA alone. For highly divergent genomes, BRAKER with protein hints may outperform RNA only mode. See the route map in Methods Mol Biol for decision trees.

Practical Workflow or Implementation Steps

The following sequence assumes you have a Linux environment, Conda or Mamba installed, and at least 32 GB RAM for a typical 500 Mb genome.

1. Prepare Input Files

• Genome assembly: in FASTA format, ideally with softmasked repeats (lowercase letters). Check for contamination using BlobTools or similar.
• RNA Seq reads: trimmed and quality controlled (use fastp or Trimmomatic). Store in FASTQ format or already aligned as BAM files. For BRAKER3 you can also use long reads (IsoSeq, ONT cDNA).
• Protein hints (optional): from closely related species, in FASTA format. Useful if RNA Seq coverage is low.

2. Align RNA Seq Reads to the Genome

Use a splice aware aligner such as HISAT2 or STAR. Example with STAR:

STAR --runThreadN 8 --genomeDir star_index --readFilesIn R1.fastq R2.fastq --outSAMtype BAM SortedByCoordinate --outFileNamePrefix sample

Index the genome first with STAR --runMode genomeGenerate. Check alignment rates (expect > 70% for a good assembly). For multiple libraries, merge BAM files.

3. Run BRAKER

BRAKER3 integrates GeneMark ET, AUGUSTUS, and TSEBRA. Basic command:

braker.pl --genome=genome.fa.masked --bam=sample.bam --softmasking --threads=8

Key parameters:

• --softmasking: use softmasked genome.
• --bam: one or more BAM files.
• --skipAllTraining: if you already have trained AUGUSTUS parameters.
• --prot: provide a protein file for protein hints (BRAKER3 long read mode requires only BAM).

Output files are placed in a braker/ directory. The main output is braker.gtf.

4. Evaluate and Iterate

After the run, inspect the number of predicted genes. For a typical insect genome expect 12,000 20,000 genes. See Annotation of 200 Insect Genomes with BRAKER for typical counts across species. If numbers are wildly off (e.g., >50,000), there may be contamination or poor masking.

5. Postprocess with Evidence Based Selection

BRAKER uses TSEBRA internally to select the best transcript models from multiple predictions. You can also run TSEBRA separately with different weights if you have additional evidence types like CAGE data.

Quality Checks

Your annotation should be evaluated with multiple metrics before use.

BUSCO completeness. Run BUSCO on the predicted protein set. Use the appropriate lineage (e.g., eukaryota_odb10, insecta_odb10). A complete BUSCO score > 90% is good, > 80% is acceptable for draft genomes. Low completeness indicates missing genes, high fragmentation suggests assembly breaks.

Gene structure statistics. Check mean exon count per gene (typically 5 10 for animals), mean intron length, and coding sequence length. Compare to published annotations for related species.

Mapping consistency. Align RNA Seq reads to the predicted transcripts (e.g., with minimap2). A high proportion of properly paired reads suggests the annotation captures expressed regions.

Manual inspection. Use IGV to view a few random loci. Look for persistent intron retention, truncated gene models at contig ends, and false fusion predictions. The Galaxy Training Network has tutorials on visualizing annotations.

Functional annotation. Assign putative functions via InterProScan or eggNOG mapper. Unexpected functional categories may reveal contamination.

Common Mistakes

Using an unmasked genome. This is the most frequent error. Repeat masking reduces false positive predictions by 30 50%. Always run RepeatModeler+RepeatMasker first.

Insufficient RNA Seq depth. One library of 50 million reads may not saturate transcript discovery. Pool at least three biologically different libraries when possible. See SRA for publicly available datasets that you can add.

Mixing nonhomologous samples. Do not combine RNA Seq from divergent species to train a single BRAKER run. This confuses the gene finder.

Ignoring assembly gaps. BRAKER will try to predict genes across gapped scaffold regions. These are often artifacts. Check the assembly for N50 and gap content.

Not checking for strand specificity. BRAKER assumes unstranded data by default. If your library is stranded, specify --stranded=rf or --stranded=fr. Mismatched strandness reduces sensitivity.

Running with default parameters without inspection. BRAKER defaults work for many genomes but may need adjustment for unusual GC content or intron length. For large introns (>50 kb) increase --max_intron_length.

Limits of Interpretation

BRAKER is a robust pipeline but it has inherent limitations that affect interpretability.

It annotates only protein coding genes. Noncoding RNAs (lncRNA, miRNA, tRNA) are not predicted. For a complete annotation you must run separate tools like Infernal for ncRNA.

It relies on evidence from RNA Seq or proteins. In regions with no expression (e.g., testis specific genes not sampled), BRAKER will miss those loci. The annotation is evidence dependent, not a complete gene set.

Pseudogenes may be included. Processed pseudogenes with near perfect ORFs can be misannotated as functional genes. Post hoc manual curation is required.

Paralog resolution is limited. In recently duplicated gene families, BRAKER may merge two copies into one model or split a single gene into two, especially if reads map ambiguously.

Transposable element contamination is reduced but not eliminated. Even with softmasking, some TE remnants can be annotated if they retain coding potential and are expressed.

Interpretation for nonmodel species. BRAKER trained parameters are species specific. Transferability to a closely related species is possible but accuracy drops. The Eukan pipeline addresses this for less studied eukaryotes.

Saturation of gene count does not guarantee correctness. You may predict the right number of genes but many could be false positives. Always validate with BUSCO and manual checks.

Frequently Asked Questions

1. Can I run BRAKER without RNA Seq data? No. BRAKER requires RNA Seq alignments or a BAM file. Without evidence, use AUGUSTUS with preexisting species parameters or Galba (a BRAKER variant that uses protein hints only).

2. How do I choose between BRAKER1 and BRAKER2? BRAKER1 uses only RNA Seq data and is simpler. BRAKER2 integrates protein hints from a database. BRAKER3 (the current version) supports long reads and both short reads and proteins. Use BRAKER3 for all new projects.

3. My annotation has very few genes. What might be wrong? Possible causes: low coverage RNA Seq reads, highly fragmented assembly, or your genome is very compact (e.g., some fungi). Check alignment rate and BUSCO completeness. In fungal genomes, TGP WEB offers an alternative automatic pipeline.

4. Do I need to run BRAKER on each chromosome separately? No. BRAKER can handle multi FASTA genome files. It will predict genes on each sequence individually. However, very small contigs (< 1 kb) will produce many incomplete predictions. It is recommended to filter out contigs below a length threshold (e.g., 1000 bp) before running.

References and Further Reading

• NCBI Bookshelf for foundational genome assembly and annotation principles.
• EMBL EBI Training with courses on RNA Seq analysis and gene prediction.
• Galaxy Training Network for hands on tutorials on BRAKER and related tools.
• Bioconductor for R packages that can postprocess BRAKER outputs (e.g., GenomicFeatures).
• NCBI Sequence Read Archive to download public RNA Seq data for your species.
• Annotation of 200 Insect Genomes with BRAKER for large scale use case and expected gene counts.
• TGP WEB server for automatic fungal genome annotation.
• Eukan pipeline for eukaryotic genomes with limited prior data.
• BRAKER route map for detailed decision criteria and workflow.
• Improved assembly of whiteleg shrimp with BRAKER showing application in a nonmodel species.
• Cercospora zeina population genomics example of BRAKER annotation in a plant pathogen.

Related Articles

• Cell Cycle
• Dna Replication
• Protein Synthesis
• Incomplete Dominance Gene
• Cell Membrane Function Biology