Antigenic annotation pipeline

This pipeline was created during my PhD at Kingston Univeristy London to predict and annotate potential vaccine candidate proteins from draft Schistosome genomes.

Problems:

• Some genomes of medically important parasites are in draft form and lack the functional annotation needed in the prediction of potential vaccine candidates.
• When faced with multiple draft genomes it can be beyond the scope of a single project to annotate multiple genomes for comparative studies.

Solution:

• Antigenic_annotation.sh predicts and annotates single protein queries from draft genomes and provides protein annotation of transmembrane regions, conserved domains and antigenic regions.
  
  
  

  Usage:  ./Antigenic_annotation.sh  /path/to/genomes/ protein_query.fasta  best_n  gene_id  translation_table
  
   1. ./Antigenic_annotation.sh = The shell script.
   2. /path/to/genomes/ = Path to directory containing target genomes.
   3.  protein_query.fasta = A single protein reference sequence in fasta format. 
   4.  best_n = The number of predictions you want exonerate to make for each genome.
   5.  gene_id = An id for naming your results files and directories e.g the name of the gene you are predicting.
   6.  translation_table = the translation table to be used for translating predictions e.g standard is 1. 
  

Part I

• Predicts best scoring coding-exon sequences from target genome/s using exonerate through the protein2genome model
• Extracts coding-exon sequence from exonerate output using extract_exonerate.py.
• Gets the highest scoring coding-exon alignment and translates it using Transeq (EMBOSS) (highest scoring is predicted using best_n 1 in exonerate)
• Uses longest_orf.py to predict the longest open reading frame.
• Stores the prediction from each species/genome in a single multi-fasta file.
  
  
  

Part II

• Aligns all proteins from the multi-fasta file (Output of Part I).
• Separates multi-fasta file into single files using mulif_to_singlef.py.
• Runs Bepipred, Tmhmm and a CDD search on each protein.
• Aligns antigenicity predition scores to the sequence alignment using align_bepipred_scores.py.
• Plots aligned antigenic scores in R using Ag_plot.R.
  
  
  

Notes on usage and requirements

This pipeline relies on a number of scripts which can be found in this repository and in the Sequence-tools repository.


To run Antigenic-annotation.sh you will need:

• extract_exonerate.py
• longest_orf.py
• align_bepipred_scores.py
• cdd_search.sh
• multif_to_singlef.py
• Ag_plot.R

All scripts mentioned above and the following programs need to be made available to $PATH:

• Exonerate (tested with v2.4.0).
• Transeq (from emboss v6.6.0.0)
• cowsay (v3.03, just for fun)
• Bepipred (tested with v1.0)
• Clustalo (tested with V1.2.4)
  
  
  

Notes on personalisation

• If you want to change the axes in the plot the easiest way is to edit them in align_bepipred_scores.py at: out.write("Sequence" + "," + "AA" + "," + "Ag" + "," + "code" + "\n" + "\n".join(final))

• At the moment the ids of the predicted sequences cannot be edited and will appear as in the target draft genome.

• Target genomes need to have the .fna extension, if you want to change this alter the extension in the line: for f in *.fna*;do in Antigenic-annotation.sh.
  
  
  

Notes on limitations and improvement:

On gene models and repeats

• It is important to know before hand the gene model of your protein so that you can manually check the predictions.
  
  
  
• Unlike in whole genome annotation, there is no step in this pipeline to mask repeat regions from the draft genome before prediction.
  
  
  
• To rectify this you can generate a species specific repeat library (using RepArk and/or Repeatmasker or dust) and softmask the genome (adding the option use --softmasktarget in the line that calls exonerate in the Antigenic-annotation.sh script).
  
  
  
• This pipeline can also be run on cDNA datasets; and by using Transdecoder you can separate out the long coding transcripts then use these to check the validity of your gene model i.e are the number of exons predicted in the expressed transcript (as predicted only from coding transcripts) the same as in the coding-exon prediction.
  
  
  

On orthology

• The prediction of orthologs can be hindered by the existence of paralogs (gene duplications). This pipeline will predict the coding-exon sequence that aligns best to a given reference sequence and cannot distinguish between orthologs and paralogs.
  
  
  

• The chance of your protein being a true ortholog can be explored after prediction:

  • Check if your coding-exon is a single copy gene (i.e in predictions, from multiple genomes of related species, is there only one hit?).
  • Making a gene tree (if this tree follows the species phylogeny then your sequence is more likely to be an ortholog).
  • Check for in frame stop codons that might indicate your sequence is a non-functional paralog. Note that longest_orf.py searches for in-frame stop codons, checking for an output txt file of these stop codons in predicted coding-exon sequences is a good place to start.
  • If available, check transcriptome data, to see if your predicted coding-exon sequence is expressed by your species.
  • Check functional annotation, i.e does the protein sequence have the same predicted functional domains as known orthologous sequences? It is generally assumed that orthologous proteins carry out the same or similar function (although, of course, there are exceptions).
    
    
    

Citation

If you use this pipeline in your work please cite as:

Eldridge, C.J.L, Majoros, G., Cook, R.T., Kidd, D., Emery, A.M., Lawton, S.P. (2021). An antigenic annotation pipeline for transmembrane proteins from draft genomes[Software]. Unpublished

Shield:

This work is licensed under a Creative Commons Attribution 4.0 International License.