AntiFP2: Antifungal Protein Prediction Toolkit

This toolkit utilizes a fine-tuned ESM2 model to classify protein sequences, with optional integration of BLAST and MERCI motif scanning. It offers three standalone prediction scripts:

1. ESM2-only classification using embeddings and a trained classifier.
2. ESM2 + BLAST + MERCI for combined evidence-based prediction.
3. MetaPipeline: Annotate contigs with Prokka and predict antifungal proteins using ESM2.

Overview

These scripts classify protein sequences into functional categories using a fine-tuned ESM2 model. Optionally, predictions can be enhanced by incorporating:

• BLAST results (from precomputed TSV files).
• MERCI motif hits (from .locate files).

Installation

Install dependencies using pip:

pip install torch==2.4.1 biopython==1.85 pandas fair-esm==2.0.0 huggingface_hub==0.30.2

Or using Conda:

conda install -c conda-forge -c bioconda -c defaults torch==2.4.1 biopython==1.85 pandas fair-esm==2.0.0 huggingface_hub==0.30.2

Clone this repository:

git clone https://github.com/patrik-ackerman/antifp2.git
cd antifp2

(Optional) Create a virtual environment:

conda env create -f environment.yml
conda activate antifp2

Scripts and Usage

1. ESM2-Only Prediction

python3 antifp2_ESM2.py --fasta input.fasta --output esm2_results.csv

Arguments:

• --fasta: Input FASTA file containing protein sequences.
• --output: Output CSV file to save predictions.
• --threshold: (Optional) Classification threshold (default: 0.5).

Output:

A CSV file with the following columns:

• ID: Sequence identifier.
• probability: Predicted probability of being an antifungal protein.
• prediction: Binary prediction (1 for antifungal, 0 for non-antifungal) based on the threshold.

Help:

python3 antifp2_ESM2.py --help

2. ESM2 + BLAST + MERCI Prediction

python3 antifp2_ESM_blast.py \
    --fasta input.fasta \
    --outdir output_directory \
    --threshold 0.5 \
    --no-cleanup

Arguments:

• --fasta: Input FASTA file containing protein sequences.
• --outdir: Output directory to save predictions and intermediate files (optional; defaults to input FASTA directory).
• --threshold: Classification threshold (default: 0.5).
• --no-cleanup: Flag to keep intermediate files like BLAST and MERCI outputs.

Description:

This script runs the fine-tuned ESM2 model to predict antifungal proteins, then adjusts predictions based on BLAST and MERCI motif hits:

• Filters sequences to keep lengths between 50 and 3000 and only standard amino acids.
• Runs MERCI motif scanning and BLASTp search against a configured database.
• Adjusts probabilities by adding +0.5 for BLAST matches to known positives and reducing -5 for BLAST matches to known negatives
• Adjusts probabilities by adding +0.5 for motif hits.
• Clips combined probabilities between 0 and 1.
• Outputs a CSV file with columns: ID, probability, blast_adjustment, motif_adjustment, combined, prediction.

Outputs:

• <prefix>_predictions.csv: Final prediction CSV file with adjusted probabilities and binary predictions.
• rejected_log.txt: Log of sequences rejected during filtering.

Example:

python antifp2_ESM_blast.py --fasta proteins.fasta --outdir results --threshold 0.6

3. MetaPipeline: Prokka + ESM2 Prediction

python3 antifp2_meta.py \
    --contigs contigs.fa \
    --outdir results_dir \
    --threshold 0.5 \
    --threads 8

Arguments:

• --contigs: Input contigs FASTA file.
• --outdir: Output directory to save prediction results and intermediate files.
• --threshold: (Optional) Classification threshold (default: 0.5).
• --threads: (Optional) Number of threads to use for Prokka (default: all available).
• --no-cleanup: (Optional) Retain Prokka intermediate files.
• --metagenome: (Optional) Enable Prokka's metagenome mode.

Workflow:

1. Prokka Annotation: Annotates input contigs to predict coding sequences.
2. Sequence Filtering: Filters out proteins: - Shorter than 50 or longer than 3000 amino acids. - Containing non-standard amino acids.
3. Prediction: Runs the fine-tuned ESM2 model on valid sequences.
4. Extraction: Saves predicted antifungal sequences to a FASTA file.

Outputs (in ``--outdir``):

• *_metapred.csv: CSV file with prediction results: - ID, probability, prediction.
• *_antifp2.fasta: FASTA file of positively predicted antifungal proteins.
• rejected_log.txt: Log of sequences excluded during filtering.

Help:

python3 antifp2_meta.py --help

Citation

If you use this tool, please cite the following resources:

• ESM2 Model:

  Lin, Z., Akin, H., Rao, R., et al. (2023). Language models of protein sequences at the scale of evolution enable accurate structure prediction. Nature, 601(7891), 277–284. DOI: https://doi.org/10.1038/s41586-021-03819-2

• Prokka:

  Seemann, T. (2014). Prokka: rapid prokaryotic genome annotation. Bioinformatics, 30(14), 2068–2069. DOI: https://doi.org/10.1093/bioinformatics/btu153 PMID: https://pubmed.ncbi.nlm.nih.gov/24642063/

• BLAST:

  Camacho, C., Coulouris, G., Avagyan, V., et al. (2009). BLAST+: architecture and applications. BMC Bioinformatics, 10, 421. DOI: https://doi.org/10.1186/1471-2105-10-421

• MERCI:

  Vens, C., Rosso, M.N., & Danchin, E.G.J. (2011). Identifying discriminative classification-based motifs in biological sequences. Bioinformatics, 27(9), 1231–1238. DOI: https://doi.org/10.1093/bioinformatics/btr110

• SeqKit:

  Shen, W., Le, S., Li, Y., & Hu, F. (2016). SeqKit: A cross-platform and ultrafast toolkit for FASTA/Q file manipulation. PLoS ONE, 11(10), e0163962. DOI: https://doi.org/10.1371/journal.pone.0163962