metapredict
metapredict is a fast, accurate, and easy-to-use predictor of intrinsic protein disorder. Given an amino acid sequence, metapredict returns a per-residue disorder score between 0 and 1 that reflects how likely each residue is to be disordered. It can also carve a sequence into discrete intrinsically disordered regions (IDRs) and folded domains, and it can generate predicted AlphaFold2 pLDDT confidence scores.
metapredict is distributed as a Python package that provides both a Python application programming interface (API) and a set of command-line tools for working with FASTA files. It is also available as a webserver for single-sequence prediction and as a Google Colab notebook for large-scale batch prediction. To install metapredict and start making predictions, see Getting Started with metapredict.
How does metapredict work?
metapredict uses a deep-learning network to generate per-residue disorder scores directly from amino acid sequence. The name reflects the original training strategy: metapredict V1 was trained to reproduce the consensus disorder assigned by a panel of independent disorder predictors (as compiled by MobiDB), so each score approximates what a collection of predictors would agree on — things got pretty “meta”, hence the name.
Later networks build on this idea. V2 combines the V1 consensus disorder signal with predicted AlphaFold2 pLDDT scores, while V3 — the current default and most accurate network — combines consensus disorder with experimental AlphaFold2 pLDDT scores trained on a much larger dataset. All three networks remain available, and V3 is a drop-in replacement for the earlier versions. A more detailed description of each network is given in Getting Started with metapredict.
Because metapredict is a lightweight sequence-based network, it is also extremely fast — it can predict disorder for entire proteomes in minutes on a CPU and in seconds on a GPU, with no length limit on the sequences it can handle.
How to cite
If you use metapredict in your work, please cite the original metapredict paper and state which version of metapredict you used (V1, V2, V2-FF, or V3):
Emenecker, R. J., Griffith, D. & Holehouse, A. S. metapredict: a fast, accurate, and easy-to-use predictor of consensus disorder and structure. Biophysical Journal 120, 4312–4319 (2021). doi:10.1016/j.bpj.2021.08.039
You may additionally cite the preprints describing later updates to metapredict; see the How to cite metapredict page for the full list.
Contents:
- Getting Started with metapredict
- metapredict from the command-line
- A quick note on selecting the metapredict network
- Predicting Disorder Scores from FASTA Files
- Predicting IDRs from a fasta file
- Predicting disorder scores from sequence
- Predicting AlphaFold2 Confidence Scores from a FASTA File
- Generate Disorder Plots from FASTA files
- Quick Disorder Graph for a Sequence
- Graph Disorder from UniProt Accession
- Graph Disorder from Protein Name
- Generate AlphaFold2 pLDDT Score Figures from FASTA
- Generate Disorder Scores for CAID from a FASTA file
- metapredict in Python
- Important updates
- Predicting Disorder
- Predicting AlphaFold2 Confidence Scores
- Predicting Disorder Domains:
- Calculating Percent Disorder:
- Graphing Disorder
- Graphing AlphaFold2 Confidence Scores
- Predicting Disorder From a .fasta File:
- Predicting AlphaFold2 confidence scores From a .fasta File
- Predict Disorder Using Uniprot ID
- Predicting AlphaFold2 Confidence Scores Using Uniprot ID
- Generating Disorder Graphs From a .fasta File:
- Generating AlphaFold2 Confidence Score Graphs from fasta files
- Generating Graphs Using UniProt ID
- Generating AlphaFold2 Confidence Score Graphs Using UniProt ID
- Predicting Disorder Domains using a Uniprot ID
- Batch prediction of disorder scores or disordered domains
- Streaming disorder predictions for very large FASTA files
- Predicting Disorder Domains from external scores
- Predicting all disorder and pLDDT scores at once
- Generating CAID-format output
- Python Module Documentation
- Acknowledgements
- HELP! Metapredict isn’t working!
- Recent changes
- How to cite metapredict