nf-core pipeline · alphafold2, colabfold, esmfold
nf-core/proteinfold
Predicts three-dimensional structures of proteins, nucleic acids and complexes using multiple folding engines.
What nf-core/proteinfold is
nf-core/proteinfold is a bioinformatics analysis pipeline designed for three-dimensional molecular structure prediction. It was originally created by researchers at the Centre for Genomic Regulation under the BovReg project and Seqera Labs, with subsequent contributions from the scientific community. The pipeline is implemented in Nextflow DSL2 and deploys individual software containers per process.
Macromolecular structure determination from sequence data often requires running different algorithmic approaches depending on the biomolecules involved. The pipeline provides a standardised framework that integrates multiple predictive tools, allowing scientists to model proteins, RNA molecules, post-translational modifications, and small-molecule ligands while directly comparing results across different models.
What the pipeline does
The workflow begins by accepting sequence data specified in a CSV samplesheet alongside FASTA files. An optional splitting step can divide large multi-entry FASTA files of monomeric sequences into manageable units. The pipeline then retrieves or verifies required reference databases and model weights for the selected prediction methods.
Structure prediction is executed using one of several strategies: combined multiple sequence alignment (MSA) searching and inference, decoupled MSA calculation (available for AlphaFold2, ColabFold, and Boltz), or direct inference from sequence using the ESMFold protein language model without an MSA. The pipeline concludes by generating individual interactive HTML visualisations, comparative HTML reports across parallel predictive runs, and overall MultiQC quality control summaries.
Key facts
- Pipeline engine
- Nextflow (DSL2)
- Supported prediction tools
- AlphaFold2, ESMFold, ColabFold, RoseTTAFold2NA, RoseTTAFold-All-Atom, AlphaFold3, HelixFold3, Boltz
- Input formats
- CSV (samplesheet), FASTA
- Output formats
- PDB/mmCIF structures, interactive HTML reports, MultiQC reports
- Container support
- Docker, Singularity
- Latest release
- 2.0.0 (2026-03-27)
What you can use it for
- Monomeric protein structure prediction
- Predicting the 3D conformation of single amino acid sequences using methods such as AlphaFold2, ColabFold, or ESMFold.
- Complex multimolecular modelling
- Modelling complexes comprising proteins, RNA sequences, post-translational modifications, and small-molecule ligands using AlphaFold3, HelixFold3, Boltz, or RoseTTAFold-All-Atom.
- Alignment-free rapid folding
- Generating structural models without performing homology searches by utilising the ESMFold protein language model.
- Genome-scale sequence batch processing
- Partitioning multi-entry monomeric FASTA files via pipeline flags to perform batch structural inference across large sequence collections.
- Constraint-guided structure generation
- Predicting multi-component molecular assemblies with user-supplied physical or spatial constraints using Boltz.
Run it here
The research agent runs nf-core/proteinfold with Nextflow on its own machine — up to 32 cores, 256 GiB of memory and 24 hours per run, which covers the test profile and real single-sample runs. It builds the samplesheet, checks it against the pipeline's schema and reads the results back. When a run is too big for one machine, it gives you the exact command for your own cluster instead.
Strengths and limits
Good at
- Integrates eight distinct folding and complex-prediction algorithms under a single unified Nextflow interface.
- Provides decoupled MSA generation, allowing alignment outputs from ColabFold to be reused directly by Boltz or ColabFold inference.
- Generates comparative interactive HTML reports to evaluate structural concordances across parallel prediction modes.
- Maintains reproducibility across computing environments through containerisation of every individual pipeline process.
Watch out for
- Reference databases required by several predictive methods are exceptionally large and may exceed user storage quotas on shared high-performance computing clusters.
- Only Boltz supports user-defined structural constraints.
- Several integrated tools, including AlphaFold2, ESMFold, and ColabFold, cannot predict structures containing RNA, post-translational modifications, or small molecules.
Questions
Can I run nf-core/proteinfold without my own cluster?
The research agent runs nf-core/proteinfold with Nextflow on its own machine — up to 32 cores, 256 GiB of memory and 24 hours per run, which covers the test profile and real single-sample runs. It builds the samplesheet, checks it against the pipeline's schema and reads the results back. When a run is too big for one machine, it gives you the exact command for your own cluster instead.
What licence is nf-core/proteinfold under?
MIT, like every nf-core pipeline.
Which integrated methods support complexes containing RNA and small molecules?
AlphaFold3, HelixFold3, Boltz, and RoseTTAFold-All-Atom support structures containing proteins, RNA, small molecules, and post-translational modifications. RoseTTAFold2NA supports proteins and RNA without small molecules.
Can predictions be made without generating multiple sequence alignments?
Yes. The pipeline supports ESMFold, which relies on a protein language model to infer three-dimensional structures directly from sequence data without requiring an MSA.
Can local copies of databases be used instead of downloading them repeatedly?
Yes. Users can supply existing database paths globally using the --db parameter, or target specific methods using flags such as --alphafold2_db, --colabfold_db, --esmfold_db, and --rosettafold_all_atom_db.
How does the pipeline handle input files containing multiple sequences?
The pipeline includes an optional --split_fasta flag that divides large files containing multiple monomeric sequences into individual entries for downstream prediction.
How to cite it
10.5281/zenodo.7437038
Written from nf-core/proteinfold's own README. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.