ESMFold inputs and parameters

You provide a single-chain protein sequence — pasted directly, fetched by UniProt ID, or imported from a previous job. The one input that most affects your result is the sequence itself: keep it within the 50–500 residue optimal range and free of non-standard residues, since length and composition drive both runtime and confidence.

Webapp input methods

The .bloom* webapp provides three input methods for ESMFold protein structure prediction:

Input type selection

Protein sequence input (default)

• Format: Raw amino acid sequence using single-letter codes
• Validation: Real-time checking for invalid characters
• Placeholder: "Enter protein sequence..."
• Multiline: 4-row text area for longer sequences

UniProt ID input

• Format: UniProt accession codes (e.g., P00533, P68871)
• Auto-fetch: Automatic sequence retrieval from UniProt database
• Validation: Real-time format checking

From previous job

• Job Output Selector: Browse completed jobs with protein sequence outputs
• Compatible tools: BLAST sequence searches, other protein analysis jobs
• Data type: Requires fasta data type output
• SMILES extraction: Automatic sequence extraction from job results

Sequence requirements

ESMFold accepts single-chain protein sequences containing only the 20 standard amino acids. Key requirements:

• Standard amino acids only: The model only accepts the canonical 20 residues (ACDEFGHIKLMNPQRSTVWY)
• No modified residues: Selenocysteine (U), pyrrolysine (O), and ambiguous codes (B, J, X, Z) are not supported
• Single chain: Multi-chain complexes and heteromers are not directly supported
• Length limits: Optimal performance for sequences between 50-500 residues; sequences up to 2,500 residues are accepted but may have reduced confidence in some regions past ~1,000 residues

Direct sequence entry

Plain amino acid sequence

• Format: Raw amino acid sequence using single-letter codes
• Validation: Real-time checking for invalid characters
• Length limits: 10-2,500 residues (optimal: 50-500)
• Case sensitivity: Automatically converted to uppercase

FASTA format

• Headers: Descriptive names for result tracking
• Multiple sequences: One sequence per FASTA entry
• Header format: >identifier description (e.g., >P12345 Hemoglobin beta chain)
• Length limits: Same as plain sequence (10-2,500 residues total)

Sequence properties display

• Length: Total amino acid count
• Molecular weight: Calculated from sequence composition
• Isoelectric point: Theoretical pI based on amino acid charges
• Composition: Amino acid frequency analysis

UniProt integration

Accession lookup

• Format: UniProt accession codes (e.g., P68871, Q9Y261)
• Validation: Real-time checking against UniProt database
• Auto-fetch: Automatic sequence retrieval and metadata loading
• Isoforms: Selection of alternative splice variants when available

Metadata integration

• Protein names: Full protein names and aliases
• Organism: Source organism information
• Function: Known or predicted protein function
• Domains: Annotated functional domains and features

Previous job integration

Compatible tools

• BLAST: Sequence search results with hit sequences
• Sequence analysis: Extracted or processed sequences
• Other folding jobs: Previously predicted structures (sequence extraction)

File handling

• Multi-sequence files: Automatic parsing and selection
• Format conversion: Handling various sequence file formats
• Provenance tracking: Maintains connection to original data source

Sequence validation and preprocessing

Character validation

• Standard amino acids only: ACDEFGHIKLMNPQRSTVWY
• Exclusion of ambiguous codes: B, J, O, U, X, Z not allowed
• Special characters: Automatic removal of whitespace, numbers, and punctuation

Sequence cleaning

• Whitespace removal: All spaces, tabs, and line breaks stripped
• Number removal: Position numbers and other numeric characters
• Case normalization: Conversion to uppercase for consistency
• Non-amino filtering: Removal of any non-amino acid characters

Length optimization

• Chunking: Automatic splitting for very long sequences (>500 residues)
• Domain splitting: Manual splitting at low-complexity regions
• Memory considerations: Guidance for optimal sequence lengths

Webapp parameters

Personalized settings toggle

• Use Personalized Settings: Checkbox to enable project-specific configuration
• Loads settings from Supabase when enabled
• Settings are keyed by esmfold_settings per project

Advanced settings (via settings panel)

Click "Advanced settings…" to access additional parameters:

Prediction quality:

• Num Recycles: Refinement iterations for structure prediction
  • Higher values improve accuracy but increase runtime
  • Configurable via global tool settings
• Chunk Size: Sequence segmentation for long proteins
  • Automatic optimization based on sequence length
  • Manual override available for specific requirements

Submit button

• Label: "Fold Protein"
• Disabled when: No sequence entered (manual) or no job output selected (previous job)
• Keyboard shortcut: Enter to submit (via EnterSubmitContainer)

Job status and ETA

• EstimatedTimeWidget: Shows estimated processing time for ESMFold jobs
• Live ETA: Real-time completion estimate during job processing
• Progress tracking: Phase-based progress indication

Output options

• PAE matrices: Predicted aligned error data (default: enabled)
• Confidence scores: Per-residue pLDDT values (default: enabled)
• Secondary structure: Predicted helices and sheets

Batch processing strategies

Single sequence jobs

• Recommended approach: One sequence per job for predictable results
• Runtime consistency: More reliable completion times
• Error isolation: Issues affect only individual sequences

Multi-sequence batches

• Short sequences: Peptides and small proteins (<100 residues)
• Aggregate limits: Total length <500 residues recommended
• Monitoring: Track individual sequence completion status

High-throughput workflows

• Queue management: .bloom*'s job queuing system handles multiple submissions
• Progress tracking: Real-time status across all batch jobs
• Result aggregation: Tools for comparing multiple predictions

File preparation guidelines

FASTA formatting best practices

• Clear identifiers: Use meaningful names for result mapping
• Standard format: >identifier|description|additional_info
• Length limits: Keep headers under 200 characters
• Special characters: Avoid pipes, commas, and other delimiters in headers

Sequence quality checks

• Reading frame: Ensure correct translation from DNA if applicable
• Start/stop codons: Proper handling of translation initiation/termination
• Splice variants: Consideration of alternative splicing effects
• Post-translational modifications: Awareness of potential PTM sites

Pre-submission validation

• External tools: Use sequence validation software if available
• Homology checking: BLAST against known structures for reference
• Domain identification: Check for known structural domains
• Unusual compositions: Flag sequences with extreme amino acid biases

Error handling and troubleshooting

Common input errors

• Invalid characters: Non-standard amino acid codes
• Length violations: Sequences outside supported range
• Format issues: Malformed FASTA or corrupted files

Validation feedback

• Real-time checking: Immediate error highlighting in webapp
• Actionable messages: Specific guidance for fixing issues
• Suggestion system: Automatic recommendations for common problems

Recovery strategies

• Sequence cleaning: Tools for removing problematic characters
• Format conversion: Utilities for fixing file format issues
• Splitting: Breaking long sequences into manageable chunks

Read next