Training Ab Initio Gene Predictors

The funannotate2 train command trains ab initio gene prediction algorithms to improve gene prediction accuracy. This step is optional but recommended for best results, especially for non-model organisms.

Basic Usage

funannotate2 train -f genome.fasta -s "Aspergillus fumigatus" -o train_results

Required Arguments

  • -f, --fasta: Input genome FASTA file

  • -s, --species: Species name (e.g., “Aspergillus fumigatus”)

  • -o, --out: Output folder name

Optional Arguments

  • -t, --training-set: Training set to use in GFF3 format

  • --strain: Strain/isolate name

  • --cpus: Number of CPUs to use (default: 2)

  • --optimize-augustus: Run Augustus mediated optimized training (not recommended) (default: False)

  • --header-len: Max length for fasta headers (default: 100)

Training Process

The train command performs the following steps:

  1. Prepare the genome:

    • The genome is loaded and quality checks are performed

    • Headers are checked to ensure they are not too long

    • Contigs are checked for non-IUPAC characters

  2. Generate a training set (if not provided):

    • BUSCOlite is used to identify conserved single-copy orthologs in the genome

    • The identified orthologs are used to create a training set of gene models

    • The training set is filtered to remove problematic gene models

  3. Split the training set:

    • The training set is split into test and train sets

    • The test set is used to evaluate the performance of the trained models

  4. Train Augustus:

    • Augustus is trained using the training set

    • The training process generates species-specific parameters for Augustus

    • The trained parameters are evaluated using the test set

  5. Train SNAP:

    • SNAP is trained using the training set

    • The training process generates species-specific parameters for SNAP

    • The trained parameters are evaluated using the test set

  6. Train GlimmerHMM:

    • GlimmerHMM is trained using the training set

    • The training process generates species-specific parameters for GlimmerHMM

    • The trained parameters are evaluated using the test set

  7. Save the trained parameters:

    • The trained parameters for all tools are saved in a JSON file

    • The JSON file can be used with the predict command

    • The trained parameters are also saved in the funannotate2 database for future use

Output Files

The train command generates the following output files in the specified output directory:

  • params.json: JSON file containing the trained parameters for all tools

  • training-models.train.gff3: The training set used for training

  • training-models.test.gff3: The test set used for evaluation

The train_misc directory contains intermediate files and detailed results from the training process:

  • augustus/: Directory containing Augustus training files

  • snap/: Directory containing SNAP training files

  • glimmerhmm/: Directory containing GlimmerHMM training files

  • busco/: Directory containing BUSCOlite results (if used to generate the training set)

Using Trained Parameters

The trained parameters can be used with the predict command in two ways:

  1. Using the params.json file:

    funannotate2 predict -f genome.fasta -o predict_results -p train_results/params.json -s "Aspergillus fumigatus"

  2. Using the species name (if the parameters have been saved in the funannotate2 database):

    funannotate2 predict -f genome.fasta -o predict_results -p aspergillus_fumigatus -s "Aspergillus fumigatus"

To save the trained parameters in the funannotate2 database, use the species command:

funannotate2 species -l train_results/params.json

This will make the trained parameters available for future use with the species name as the identifier.