FunGAP: Fungal Genome Annotation Pipeline v1.0.1
Last updated: Sep 4, 2019
FunGAP is freely available for academic use. For the commerical use or licensing of FunGAP, please contact In-Geol Choi (igchoi (at) korea.ac.kr). Please, cite the following reference
Reference: Byoungnam Min, Igor V Grigoriev, and In-Geol Choi, FunGAP: Fungal Genome Annotation Pipeline using evidence-based gene model evaluation (2017), Bioinformatics, Volume 33, Issue 18, Pages 2936–2937, https://doi.org/10.1093/bioinformatics/btx353
INSTALL.md and USAGE.md for installation and usage instruction, or you can go to wiki tab for the same.FunGAP performs gene prediction on given genome assembly and RNA-seq reads. See
- FunGAP INPUT & OUTPUT
- Pipeline description
FunGAP INPUT & OUTPUT
--output_dir | Output directory --trans_read_files | Illumina paired-end mRNA reads files (FASTQ) --genome_assembly | Genome assembly file (FASTA) --augustus_species | Augustus --species argument --sister_proteome | Protein database (FASTA) --num_cores | Number of CPU cores to be used
fungap_output.gff3 | Tab-delimited genomic feature format fungap_output_prot.faa | Translated protein sequences fungap_output_result_summary.html | Summary of FunGAP results
Step1: Preprocessing of input data
In preprocessing step, FunGAP masks repeat regions in genome assembly (input data 1) and assembles mRNA reads into transcript contigs (input data 2).
Repeat masking is a crucial step in eukaryotic gene prediction because genomic regions, such as transposon repeats, often make false alignments and interfere with gene prediction. FunGAP employs a repeat masking procedure embedded in the Maker pipeline along with a genome-specific repeat library built by RepeatModeler (http://www.repeatmasker.org/RepeatModeler.html).
Assembly of mRNA reads
User-provided mRNA reads are assembled by the Trinity program. A BAM-format file for genome-guided assembly is generated by a Hisat2 read aligner and Samtools format converter (SAM file to sorted BAM file). An optional parameter
--jaccard_clip in Trinity is used for fungal transcript assembly because high gene density leads to UTR overlap in the assembly. This option helps avoid fusion of neighbor transcripts. The maximum intron length is set to 2000 bp with the
--max-intronlen option in Hisat2.
Step2: Gene prediction
FunGAP uses three gene prediction tools: Augustus, Braker, and Maker. The outcomes of predictions are stored in GFF3 and FASTA files for the next set of evidence score calculations.
Maker and default parameters used by FunGAP
FunGAP runs Maker four times with iterative SNAP gene model training, as previously described. FunGAP uses the correct_est_fusion option to correct fusion of neighbor transcripts in mRNA assembly due to the above-mentioned high gene density of fungal genomes. Maximum intron length is set to 5000 bp with the
split_hit option. Single-exon genes longer than 50 amino acids are predicted by setting the single_exon and single_length options.
Augustus and default parameters used by FunGAP
FunGAP runs Augustus with the augustus_species parameter specified by a user. The option
--softmasking is turned on as repeat-masking generates soft-masked assembly. To allow overlapping CDS predictions, FunGAP turns on the
--singlestrand option. The output is GFF3, and translated protein sequences are generated in FASTA by a simple parsing script.
Braker and default parameters used by FunGAP
Braker performs unsupervised RNA sequencing-based genome annotation using GeneMark-ET and Augustus. The option
--softmasking is turned on as repeat-masking generates soft-masked assembly. The input file for Braker is the mRNA reads alignment formatted in a BAM file produced in the preprocessing step.
Step3: Gene model evaluation and filtration
In the previous step, three gene predictors generated a set of predicted genes (designate as “gene models” hereafter). FunGAP produces “non-overlapping” coding sequences by evaluating all gene models and retaining only best-scored models. The evaluation is performed by three tools: BLASTp, Benchmarking Universal Single-Copy Orthologs (BUSCO), and InterProScan. Bit scores from alignments are multiplied by length coverage because longer gene models have more chances to get higher alignment scores. The sum of three scaled bit scores becomes the evidence score for each gene model. Finally, the filtration produces a final set of gene models.
Sequence similarity with genes in phylogenetically close genomes can be an evidence for predicted genes being actual genes. Users provide the proteome of phylogenetically related organisms with the
--sister_proteome argument. For convenience, FunGAP provides a script, download_sister_orgs.py, which downloads protein sequences from NCBI for a given taxon. To reduce computing time, FunGAP integrates the gene models from three gene predictions, and removes identical gene models to make nonredundant gene models.
BUSCO provides hidden Markov models for single-copy orthologs conserved in all fungal genomes. Evidence scores for BUSCO are calculated by multiplying “full sequence scores” in hmmer output and length coverage [min (query length, target length)/max (query length, target length)].
InterProScan (Pfam domain prediction)
Pfam provides a database of manually curated protein families. We assume that gene models annotated with a Pfam domain are more likely to be an actual gene. Evidence scores for Pfam are directly provided by the hmmer3-match score in the XML output of InterProScan (-f XML option). For multiple domains in one gene model, the sum of the scores is used.
Sequence similarity with assembled transcriptome can give the direct evidence for reliability of predicted genes. FunGAP runs BLASTn for each predicted gene against Trinity-assembled transcripts. Length coverage is also considered.
Three bit scores gained from the above four sources are summed to provide evidence scores for each gene model. The equation of this scoring function is as follows:
Evidence score (gene model) = BLASTp_score*cov(query)*cov(target) + BUSCO_score + Pfam_scores + BLASTn_score*cov(query)*cov(target)
In the filtration process, FunGAP finds “gene blocks” defined as a set of gene models that overlap with at least one base pair. FunGAP gets all combinations of gene models in a gene block and calculates the sum of the evidence scores. Gene models in the block with the highest evidence score are selected as final genes of that region. Short coding sequence overlap (less than 10% of coding sequence length) is allowed.
- Project principal investigator: Prof. In-Geol Choi, CSBL at Korea University
- Contact (email-address): igchoi at korea.ac.kr or mbnmbn00 at korea.ac.kr (or mbnmbn00 at gmail.com)
If you have any problem to install or run, please don't hesistate to contact us. We will help you as much as we can. Any input from users will help to build more robust pipeline.