# Protein structure prediction

Fold a protein sequence online with AlphaFold2 or ESMFold, predict complexes with ligands or nucleic acids using Boltz-2 or Chai-1, and see which regions each model is confident about.

[Run AlphaFold2](https://proteiniq.io/app/alphafold-2)[Open comparison workflow](https://proteiniq.io/app/workflows/templates?template=protein-structure-prediction-method-panel)[Contact sales](https://proteiniq.io/contact)

## More protein structure prediction workflows

## Tools for protein structure prediction

Use these methods to prepare inputs, run the core analysis, inspect outputs, and validate the evidence described in this workflow.

[![AlphaFold2](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Falphafold-2.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### AlphaFold2 General monomer and multimer structure prediction](https://proteiniq.io/app/alphafold-2)[![ESMFold](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fesmfold.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### ESMFold Fast single-sequence structure prediction](https://proteiniq.io/app/esmfold)[![ESMFold2](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fesmfold-2.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### ESMFold2 Single- and multi-chain language-model folding](https://proteiniq.io/app/esmfold-2)[![Boltz-2](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fboltz-2.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### Boltz-2 All-atom biomolecular structure prediction](https://proteiniq.io/app/boltz-2)[![Chai-1](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fchai-1.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### Chai-1 Multi-component biomolecular prediction](https://proteiniq.io/app/chai-1)[![OpenFold3](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fopenfold-3.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### OpenFold3 Open all-atom structure prediction](https://proteiniq.io/app/openfold-3)[![Protenix](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fprotenix.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### Protenix Protein and biomolecular complex prediction](https://proteiniq.io/app/protenix)[![RosettaFold3](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Frosettafold3.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### RosettaFold3 Protein and complex structure prediction](https://proteiniq.io/app/rosettafold3)[![ABodyBuilder3](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fabodybuilder3.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### ABodyBuilder3 Antibody-specialized structure prediction](https://proteiniq.io/app/abodybuilder3)[![HighFold](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Falphafold-2.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### HighFold Cyclic peptide structure prediction](https://proteiniq.io/app/highfold)[![MolProbity](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fmolprobity.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### MolProbity All-atom model-quality review](https://proteiniq.io/app/molprobity)[![USAlign](https://proteiniq.io/_next/image?url=%2Fimages%2Ftools%2Fusalign.webp&w=96&q=75&dpl=dpl_D9qEtqGGob9dunzQod4hmU4UryfW) ### USAlign Structural comparison and alignment](https://proteiniq.io/app/usalign)

## What is protein structure prediction?

Protein structure prediction is the computational process of estimating a protein’s local secondary structure or three-dimensional shape from its amino-acid sequence. Methods can incorporate evolutionary relationships, structural templates, or learned molecular representations; the appropriate model depends on whether the question concerns local states, a monomer, an assembly, or a specialized molecular class.

Protein structure prediction is not one interchangeable task. Homology modeling transfers information from related experimental structures, single-sequence models infer a fold without a user-supplied alignment, and complex predictors jointly model multiple chains. Antibodies, peptides, and membrane proteins require more specific input and interpretation.

A coordinate file is a hypothesis, not a biological conclusion. Confidence can vary by residue, domain, interface, conformational state, and similarity to the evidence a predictor learned from; retain native confidence and compare predictions where the decision warrants it.

### When to use protein structure prediction

Choose a predictor according to the structural object you need to study:

- Generate a structural hypothesis for construct design, interpretation, docking, engineering, or experiment planning.
- Use complex-, antibody-, peptide-, or membrane-aware interpretation instead of treating every sequence as a soluble monomer.
- Run complementary predictors from consistent inputs and retain structures, confidence, settings, and files.

### Strengths and limitations

Structure predictors can turn a sequence into coordinates that make residue environments, domain organization and possible interfaces inspectable. Comparing complementary models is useful when their confidence and disagreements remain attached to each structure, especially before using a model for docking or design.

Confidence varies across domains, loops, termini and interfaces. Ligands, membranes, partners, modifications and conformational states may be absent from the input, so a convincing monomer model can still represent the wrong biological state. Coordinates alone cannot establish mechanism, binding or function.

## Which protein structure prediction tool to use

Start from what you are folding: one chain, a complex, a complex with a ligand or nucleic acid, or a specialized molecule such as an antibody. Every tool below runs online from a pasted sequence or FASTA file and quotes its credit cost before it starts. With default settings, a prediction of the 76-residue ubiquitin sequence costs 50 credits with ESMFold and 100 with AlphaFold2.

| What you want to do | Tool | What you get back |
| --- | --- | --- |
| Fold one protein quickly from its sequence | [ESMFold](https://proteiniq.io/app/esmfold) | A structure with per-residue pLDDT confidence |
| Fold a protein or protein complex with an MSA | [AlphaFold2](https://proteiniq.io/app/alphafold-2) | Ranked models with pLDDT, pTM, ipTM and PAE |
| Predict a complex with a ligand, DNA or RNA | [Boltz-2](https://proteiniq.io/app/boltz-2) | Complex structures, confidence metrics and a ligand affinity estimate |
| Get a second all-atom prediction to compare | [Chai-1](https://proteiniq.io/app/chai-1) | Ranked predictions with pLDDT, ipTM and clash indicators |
| Predict an antibody, nanobody or T-cell receptor | [ImmuneBuilder](https://proteiniq.io/app/immunebuilder) | A refined structure chosen from a model ensemble |
| Predict a cyclic peptide | [HighFold](https://proteiniq.io/app/highfold) | Cyclic peptide models ranked by pLDDT |
| Estimate helix and sheet regions from sequence alone | [Chou-Fasman](https://proteiniq.io/app/chou-fasman) | Per-residue helix, sheet and turn assignments |
| Check the geometry of a model | [MolProbity](https://proteiniq.io/app/molprobity) | Clashscore, Ramachandran and rotamer outliers |

ESMFold is the fastest way to see whether a sequence folds confidently, because it needs no alignment. AlphaFold2 is slower but usually more accurate when the protein has many homologs, and it reports interface confidence for complexes. When two predictors disagree about a domain or an interface, treat that region as uncertain rather than picking the better-looking model.

## Types of protein structure prediction

The maintained prediction types reflect distinct researcher workflows. [Homology modeling](https://proteiniq.io/use-cases/homology-modeling) starts from a related experimental structure and a reviewed alignment; ProteinIQ's current workflow reviews an externally built model. [Protein secondary structure prediction](https://proteiniq.io/use-cases/protein-secondary-structure-prediction) concerns local helix, sheet and turn states. [Single-sequence protein structure prediction](https://proteiniq.io/use-cases/single-sequence-protein-structure-prediction) builds a fold without a user-supplied MSA.

[Protein complex structure prediction](https://proteiniq.io/use-cases/protein-complex-structure-prediction) adds chain composition and interface uncertainty. [Antibody structure prediction](https://proteiniq.io/use-cases/antibody-structure-prediction) emphasizes paired variable domains and CDR loops. [Peptide structure prediction](https://proteiniq.io/use-cases/peptide-structure-prediction) needs attention to cyclization and conformational ensembles, while [transmembrane protein structure prediction](https://proteiniq.io/use-cases/transmembrane-protein-structure-prediction) needs membrane topology and lipid context.

Threading and fold recognition remain important terms, but they are not presented as prediction spokes because ProteinIQ does not provide dedicated engines for them. A general predictor would not faithfully substitute for those methods.

## Applications of protein structure prediction

A monomer model can help locate conserved residues, choose construct boundaries or suggest a fold to test experimentally. A complex model can suggest interface contacts for cross-linking or mutagenesis. In both cases, local confidence and disagreement between predictors help decide where the coordinates are useful.

Specialized models support different decisions: antibody loops guide paratope hypotheses, peptide conformers suggest structural experiments and membrane-protein models help test topology. These uses require the biological state to be represented correctly; an isolated sequence does not encode every partner, modification or environmental constraint.

## How to predict protein structure online

A defensible run preserves the exact sequence or assembly and every model's native confidence. The featured panel compares four sequence-to-structure predictors for one protein input; use a specialist spoke for complexes, antibodies, peptides or membrane context.

1. **Prepare the sequence.** Confirm construct boundaries, unusual residues and whether a monomer model matches the biological question.
2. **Run an evolutionary-input branch.** [AlphaFold2](https://proteiniq.io/app/alphafold-2) produces a structure with [pLDDT](https://proteiniq.io/guides/plddt) and related confidence outputs; retain any alignment provenance.
3. **Compare sequence-derived models.** [ESMFold](https://proteiniq.io/app/esmfold) predicts from the same protein sequence without a supplied MSA.
4. **Review additional predictors.** Inspect [OpenFold-3](https://proteiniq.io/app/openfold-3) outputs and the separate [Boltz-2](https://proteiniq.io/app/boltz-2) branch without merging model-specific scores.
5. **Inspect disagreement.** Compare domain arrangement, low-confidence regions and geometry, then export all models and settings for follow-up.

## How to interpret protein structure prediction results

Local confidence is a model-specific estimate, not an experimental resolution. A well formed domain can still have uncertain placement relative to another domain or partner. Compare structures after alignment and keep each method's confidence label attached to its model.

The result can prioritize construct design and structural experiments. It does not confirm binding, stability, activity, safety or function.

## How the featured workflow works

The hub workflow compares general sequence-to-structure models; each specialist page provides a workflow matched to that prediction type.

1. **Define the question.** Confirm that a general monomer comparison matches the biological system.
2. **Prepare the sequence.** Use one clean, ungapped protein sequence with the intended construct boundaries.
3. **Run four predictors.** Submit the same sequence to [AlphaFold2](https://proteiniq.io/app/alphafold-2), ESMFold, OpenFold3, and Boltz-2.
4. **Compare predictions.** Review local confidence, domain arrangement, geometry, and structural agreement.
5. **Validate and export.** Select models in relation to the downstream question and export provenance.

## Inputs and outputs for the featured workflow

Check formats before running, then inspect and download the result from every workflow step.

### Inputs

#### Protein sequence

`FASTA``TXT`

One protein sequence for the general comparison, or the specialist inputs described on each spoke.

### Outputs

#### Structures and confidence

`PDB``mmCIF``JSON``CSV`

Model-native structures, rankings, confidence scores, PAE where supported, logs, and downloads.

## Frequently asked questions

AI answers from our docs. No sign-in needed.

## Run your first analysis today

Start with 200 free credits and no installs. Upgrade when your project needs more compute.

[Start for free](https://proteiniq.io/sign-up)[View pricing](https://proteiniq.io/pricing)