---
title: ESMFold quickstart
description: Fold a sequence with ESMFold in 10-60 seconds — paste it or fetch it by UniProt ID, keep 3 recycles, then read the pLDDT coloring and download PDB and PAE.
---

You want a 3D structure for an amino acid sequence — no alignment, no database lookup — and [ESMFold](/docs/esmfold) delivers it in a few steps: paste the sequence (or fetch it by UniProt ID), keep the defaults, submit, and read the confidence-colored model, usually within 10–60 seconds.

## Quick protein folding with .bloom* platform

ESMFold is a protein structure prediction method that leverages the ESM-2 protein language model to predict 3D structures directly from amino acid sequences. Unlike MSA-based methods, ESMFold does not require multiple sequence alignments, which makes it significantly faster at a measured cost in accuracy — see [how ESMFold results compare to the bench](/docs/esmfold/translatability) before acting on a prediction.

## Step-by-step folding workflow

### 1. Access the ESMFold tool

Navigate to the **Protein Folding** tool in .bloom*'s interactive workspace. The tool provides guided input validation and real-time feedback.

### 2. Prepare your protein sequence

**Option A: Direct sequence entry (fastest)**
- Click **Sequence** in the input section
- Paste your amino acid sequence (e.g., "MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYR")
- .bloom* automatically validates the sequence and shows properties
- For FASTA format, include a header line starting with ">"

**Option B: UniProt lookup**
- Click **UniProt ID** in the input section
- Enter a UniProt accession (e.g., "P68871" for hemoglobin subunit beta)
- .bloom* fetches the canonical sequence automatically
- Review sequence details and select isoforms if available

**Option C: From previous jobs**
- Click **From Previous Job** in the input section
- Select sequences from BLAST searches, sequence analysis, or other tools
- .bloom* handles multi-sequence files and format conversion
- Maintains provenance tracking to original data source

### 3. Configure folding parameters

**Basic settings (recommended):**
- Keep default parameters for first runs
- Job name: Enter descriptive name or leave blank for auto-generation

**Advanced settings (via gear icon):**
- **Num recycles**: Default is 3; increase up to 10 for marginally better accuracy at the cost of runtime
- **Chunk size**: Adjust for very long sequences (>500 residues)
- **CPU offloading**: Enable for memory management on large proteins

### 4. Submit and monitor the job

- Click **Run Protein Folding** to start prediction
- .bloom* shows real-time progress through folding phases
- Typical completion: 10-60 seconds for average proteins (50-200 residues)
- Receive notifications when complete (if enabled in settings)

### 5. Review and analyze results

**Interactive 3D visualization:**
- Structure loads automatically with confidence coloring
- Green: High confidence (pLDDT > 90)
- Yellow: Medium confidence (70-90)
- Red: Low confidence (< 70)
- Rotate, zoom, and inspect different regions

**Quality assessment:**
- Check overall confidence score in the results panel
- Examine per-residue confidence distribution
- Review predicted secondary structure elements
- Look for domain boundaries and flexible regions

**Download options:**
- **PDB file**: 3D coordinates for external tools
- **PAE matrix**: Predicted aligned error data
- **Metadata**: Job parameters and runtime information
- **Confidence data**: Per-residue pLDDT scores

## Common folding scenarios

### Single protein analysis
1. Enter sequence via direct input or UniProt ID
2. Use default parameters for initial assessment
3. Review confidence coloring in 3D viewer
4. Download PDB for further analysis or docking

### High-throughput screening
1. Prepare FASTA file with multiple sequences
2. Submit as batch job (one sequence per job recommended)
3. Use consistent parameters across all predictions
4. Compare confidence scores to prioritize targets

### Domain analysis
1. Submit full-length sequence first
2. Identify low-confidence linker regions
3. Split into individual domains for separate folding
4. Compare domain arrangements and interfaces

### Comparative modeling
1. Predict structures for wild-type and mutant sequences
2. Use same parameters for fair comparison
3. Analyze structural differences in 3D viewer
4. Quantify confidence changes in mutated regions

## Quality control checklist

### Sequence preparation
- [ ] Only standard amino acids (ACDEFGHIKLMNPQRSTVWY)
- [ ] No ambiguous characters (BZJOUX)
- [ ] Reasonable length (10-2,500 residues accepted; 50-500 optimal)
- [ ] No unexpected numbers or special characters
- [ ] Correct reading frame (if from DNA translation)

### Job configuration
- [ ] Appropriate sequence length for available memory
- [ ] Sufficient num_recycles for desired accuracy
- [ ] Reasonable chunk_size for long sequences
- [ ] Clear job naming for result organization

### Result validation
- [ ] Overall pLDDT score &gt; 70 for reliable structures
- [ ] High-confidence cores (>90 pLDDT) present
- [ ] Secondary structure elements properly formed
- [ ] No major steric clashes or structural artifacts

## Troubleshooting common issues

### Invalid sequence errors
- Check for non-standard amino acid codes
- Remove whitespace, numbers, and special characters
- Verify sequence is in single-letter amino acid code
- Use sequence validation tools if unsure

### Low confidence predictions
- Try increasing num_recycles (default 3, up to 10)
- Check if sequence contains unusual amino acid compositions
- Consider if protein may have intrinsically disordered regions
- Split long sequences into individual domains

### Memory or runtime issues
- Reduce sequence length or split into domains
- Enable CPU offloading in advanced settings
- Use smaller chunk_size for very long proteins
- Consider submitting shorter sequences separately

### Unexpected structure
- Verify sequence orientation (N- to C-terminus)
- Check for sequence truncation or mutations
- Compare with known homologous structures
- Consider experimental validation for critical applications

## Next steps after folding

**Structure refinement:**
- Run molecular dynamics simulations for equilibration
- Apply energy minimization to reduce steric clashes
- Consider side-chain optimization for better packing

**Functional analysis:**
- Use DiffDock for ligand binding predictions
- Analyze protein-protein interaction interfaces
- Identify functional sites and binding pockets

**Comparative studies:**
- Align with experimental structures (if available)
- Compare mutant vs. wild-type structures
- Assess stability changes and folding thermodynamics

**Integration with workflows:**
- Feed structures directly to docking tools
- Use for virtual screening campaigns
- Export for publication-quality figures

## Read next