cgeb2001's DokuWiki!

You've loaded an old revision of the document! If you save it, you will create a new version with this data.

====== Gene prediction with the Funannotate pipeline ======

Joran Martijn (December 2022)

Funannotate is a genome prediction, annotation, and comparison software package. It was originally written to annotate fungal genomes (small eukaryotes ~ 30 Mb genomes), but has evolved over time to accommodate larger genomes.

In my experience it seems to do quite a lot better in predicting gene models than the Braker2 pipeline with //Ergobibamus cyprionides// . In addition to gene prediction, it can also facilitate functional annotation (hence the name FUNctional - ANNOTATE)

An additional advantage is that it has the capacity to prepare all the files necessary for a NCBI GenBank submission.

Official documentation can be found in their [[https://funannotate.readthedocs.io/en/latest/index.html|ReadTheDocs]]

===== Clean, sort, mask and train =====

If you have a genome assembly in plain FASTA format ready, as well as some RNAseq data, you can follow along with the Funannotate tutorial described [[https://funannotate.readthedocs.io/en/latest/tutorials.html|here]].

Here I've adapted those commands so they work with our cluster Perun. Note that all these code snippets below are also represented in the [[https://github.com/RogerLab/gospel_of_andrew|Gospel Of Andrew]]

==== Clean ====

The first step is ''funannotate clean''. This algorithm attempts to find and delete short contigs which are 'repetitive', that is they are already fully represented in a larger contig (≥ 95% sequence similarity and  ≥ 95% sequence coverage overlap).

<code>
#!/bin/bash
#$ -S /bin/bash
#$ -cwd
#$ -m bea
#$ -M joran.martijn@dal.ca
#$ -pe threaded 1

source activate funannotate

# input
GENOME='ergo_cyp_genome.fasta'

# run funannotate
funannotate clean \
    --input $GENOME \
    --out ${GENOME/fasta/clean.fasta}
</code>

==== Sort ====

The second step is ''funannotate sort''. It will sort your contigs from longest-to-shortest within the FASTA file and rename the contig headers. Apparently NCBI limits the number of characters for FASTA headers to 16, and AUGUSTUS can also have issues with longer contig names. If you do not wish to rename your sequences, you can alternatively pass ''--simplify'' or ''-s'', which simply splits headers at the first space. (So if your headers do not have a space to begin with, they will remain the same).

<code>
#!/bin/bash
#$ -S /bin/bash
#$ -cwd
#$ -m bea
#$ -M joran.martijn@dal.ca
#$ -pe threaded 1

source activate funannotate

# input
GENOME='ergo_cyp_genome.fasta'

## sort by length
funannotate sort \
    --input ${GENOME/fasta/clean.fasta} \
    --out ${GENOME/fasta/sort.fasta} \
    --simplify
</code>

==== Mask ====

The third step is to mask your genome assembly. This is analogous to the RepeatMasking step of the Braker2 pipeline. The ''funannotate mask'' command default is to run simple masking using ''tantan''. The script is a wrapper for RepeatModeler and RepeatMasker, however you can use any external program to softmask your assembly. Softmasking is where repeats are represented by lowercase letters and all non-repetitive regions are uppercase letters.

<code>
#!/bin/bash
#$ -S /bin/bash
#$ -cwd
#$ -m bea
#$ -M joran.martijn@dal.ca
#$ -pe threaded 10

source activate funannotate

# input
GENOME='ergo_cyp_genome.fasta'
THREADS=10

# soft-mask with tantan
funannotate mask \
    --input ${GENOME/fasta/sort.fasta} \
    --out ${GENOME/fasta/mask.fasta} \
    --method tantan \
    --cpus $THREADS
</code>

==== Train ====

In this ''funannotate train'' step, you will map your RNAseq data against the sorted, cleaned and masked assembly with **Hisat2**, do a genome-guided transcriptome assembly with **Trinity** and another transcriptome assembly with **PASA**, all in a single step:

<code>
#!/bin/bash
#$ -S /bin/bash
#$ -cwd
#$ -m bea
#$ -q 768G
#$ -M joran.martijn@dal.ca
#$ -pe threaded 40

source activate funannotate

# input
GENOME='ergo_cyp_genome.fasta'
RNASEQ_PRD_FW='eb_rna_fw.prd.fastq.gz'
RNASEQ_PRD_RV='eb_rna_rv.prd.fastq.gz'
GENUS='Ergobibamus'
SPECIES='cyprinoides'
STRAIN='CL'
THREADS=40

## align rnaseq data, run trinity and pasa
funannotate train \
    --input ${GENOME/fasta/mask.fasta} \
    --out funannotate_out \
    --left $RNASEQ_PRD_FW \
    --right $RNASEQ_PRD_RV \
    --stranded RF --jaccard_clip \
    --max_intronlen 3000 \
    --species "$GENUS $SPECIES" \
    --strain $STRAIN \
    --cpus $THREADS  \
    --no_normalize_reads \
    --no_trimmomatic \
    --memory 100G

</code>

This step requires quite a bit of memory, and hence we're asking for the 768G RAM node of Perun. In the above snippet we assume that the RNAseq reads have already been treated with **Trimmomatic**. If they have not, and they are still raw reads, funannotate should be able to do the trimmomatic cleaning for you (though I have not tested this). We are also running with ''--jaccard_clip'' , which is apparently used when you are expecting a high density genome (like that of yeast, for example)

NOTE also that this step generates the `funannotate_out` output directory, which can be used as an input argument in future funannotate jobs.

==== Predict ====

<code>
#!/bin/bash
#$ -S /bin/bash
#$ -cwd
#$ -m bea
#$ -M joran.martijn@dal.ca
#$ -pe threaded 40

source activate funannotate

# input
GENOME='ergo_cyp_genome.fasta'
GENUS='Ergobibamus'
SPECIES='cyprinoides'
STRAIN='CL'
THREADS=40

export GENEMARK_PATH=/scratch2/software/gmes_linux_64-aug-2020/

# run funannotate
funannotate predict \
    --input ${GENOME/fasta/mask.fasta} \
    --out funannotate_out \
    --species "$GENUS $SPECIES" \
    --strain $STRAIN \
    --cpus $THREADS
</code>

Many of the required inputs do not have to be explicitly specified, since they have been generated in the previous ''funannotate train'' step and they are available in the ''funannotate_out'' directory.

Funannotate (according to log files) uses **AUGUSTUS**, **CodingQuarry**, **GeneMark-ES**, **GlimmerHMM** and **SNAP** as main gene model predictors. Note that this is more than used by Braker2, which mainly uses AUGUSTUS and GeneMark. This may be a reason why the Funannotate prediction pipeline seems to do better than Braker2 for //Ergobibamus//.

Each program uses a different source for training its algorithms. Each of these have been generated in the previous step.

As with Braker2 pipeline, funannotate uses the **EVidenceModeler (EVM)** to compile the best overall gene models from the above set of gene predictions. For //Ergobibamus// the weights looked something like this:

<code>
  Source         Weight   Count
  Augustus       1        4462
  Augustus HiQ   2        660
  CodingQuarry   2        6325
  GeneMark       1        5433
  GlimmerHMM     1        5830
  pasa           6        6470
  snap           1        3924
  Total          -        33104
</code>

In some of the final steps, funannotate calls upon ''tRNAscan'' (I'm guessing the SE version) to predict tRNA genes. Then at the very end, if you have specified it to, it can create GenBank format-compliant files for you to submit to GenBank. Awesome!

Please fill all the letters into the box to prove you're human. A H W G᠎ I Please keep this field empty:

Edit summary

Note: By editing this page you agree to license your content under the following license: CC Attribution-Noncommercial-Share Alike 4.0 International