Skip to main content
Protein structure prediction is the foundation of structure-based drug design. LiteFold provides access to state-of-the-art AI models that predict protein structures with near-experimental accuracy.

​
Overview

LiteFold’s structure prediction capabilities include:
  • Single-chain protein prediction: From sequence to 3D structure
  • Multi-chain complexes: Protein-protein interactions
  • Protein-ligand complexes: Direct prediction of bound structures
  • Mutation analysis: Predict effects of variants
  • Batch processing: High-throughput predictions

​
Available Models

AlphaFold2

Industry-leading accuracy. Best for single proteins and protein complexes.

ESMFold

Ultra-fast predictions using protein language models. Great for large-scale screens.

RoseTTAFold

Generates diverse conformations. Useful for flexible proteins.

DiffDock

Predicts protein-ligand complex structures directly from sequence and SMILES.

​
Quick Start

1

Input Your Sequence

Paste a protein sequence in FASTA format, upload a file, or search by UniProt ID.
>MyProtein
MKTFIFLALLGAAVAFPVDDDDKIVGGYTCAANSIPYQVSLNSGSHFCGGSLINSQWVVSAAHCYKS
2

Select Model

Choose your prediction model based on your needs:
  • AlphaFold2: Most accurate (5-15 min)
  • ESMFold: Fastest (< 1 min)
  • RoseTTAFold: Multiple conformations (10-20 min)
3

Configure Options

  • Enable structure relaxation (recommended)
  • Set confidence threshold
  • Specify known domains or regions
4

Run Prediction

Click “Predict Structure” and wait for results. You’ll receive a notification when complete.

​
Understanding Results

​
Confidence Scores

pLDDT (predicted Local Distance Difference Test) Measures per-residue confidence (0-100):
  • > 90: Very high confidence (blue in viewer)
  • 70-90: Generally reliable (light blue/green)
  • 50-70: Low confidence (yellow/orange)
  • < 50: Very low confidence (red)
PAE (Predicted Aligned Error) Measures relative position confidence between residue pairs. Low PAE indicates confident relative positioning.

​
3D Visualization

The interactive viewer lets you:
  • Rotate and zoom the structure
  • Color by confidence, secondary structure, or property
  • Measure distances and angles
  • Highlight specific residues
  • Create publication-ready images

​
Downloads

Export your results:
  • PDB file: Standard protein structure format
  • mmCIF file: Enhanced format with metadata
  • Images: High-resolution figures
  • Report PDF: Summary with confidence metrics

​
Advanced Features

​
Multi-Chain Prediction

Predict protein complexes and protein-protein interactions. 
>ChainA
MKTFIFLALLGAAVAFPVD...
>ChainB
GSHFCGGSLINSQWVVSAAHCYKS...
LiteFold automatically detects multiple chains and predicts the complex structure with interface confidence scores.

​
Template-Based Modeling

Provide a template structure to guide prediction:
  1. Upload or select a template PDB
  2. LiteFold aligns your sequence to the template
  3. Prediction uses template constraints for improved accuracy
Useful for homology modeling and comparative analysis.

​
Mutation Scanning

Predict effects of mutations systematically: 
Position 123: Try all 20 amino acids
→ Predicts 20 structures
→ Ranks by stability and confidence
→ Identifies critical residues

​
Ensemble Generation

Generate multiple diverse conformations:
  • Sample different folding pathways
  • Explore conformational space
  • Identify flexible regions
  • Use for ensemble docking

​
Use Cases

  • Predict structure of novel disease targets
  • Identify druggable binding pockets
  • Assess structural uniqueness
  • Prioritize targets for screening campaigns
  • Map potential binding pockets
  • Characterize pocket properties (volume, hydrophobicity)
  • Compare to known drug targets
  • Guide fragment screening
  • Model drug-resistant mutations
  • Predict impact on inhibitor binding
  • Design second-generation inhibitors
  • Proactively address resistance
  • Design stabilizing mutations
  • Engineer binding specificity
  • Optimize expression and solubility
  • Validate designs computationally

​
Best Practices

Start with AlphaFold2: For most applications, AlphaFold2 provides the best balance of accuracy and speed.
Check confidence scores: Focus on regions with pLDDT > 70 for drug design. Low confidence regions may be disordered or require experimental validation.
Use ensembles for flexible proteins: If you suspect conformational flexibility, generate an ensemble and use all conformations for docking.
Limitations: Current models struggle with:
  • Intrinsically disordered regions
  • Large conformational changes
  • Novel folds with no homologs
  • Some membrane proteins
Always validate critical findings experimentally.

​
Batch Processing

Process multiple sequences in parallel:
  1. Upload a multi-FASTA file or CSV with sequences
  2. Select prediction parameters
  3. LiteFold queues all predictions
  4. Results are organized by sequence ID
  5. Download all structures or analysis summaries
Perfect for:
  • Genome-wide structural analysis
  • Protein family studies
  • High-throughput target assessment

​
Integration with Other Tools

Structure predictions automatically flow to:
  • Docking: Use predicted structure as receptor
  • Molecular Dynamics: Validate and refine structure
  • De Novo Design: Use as template for molecule generation
  • Rosalind AI: Ask questions about the structure

​
Performance

Typical prediction times:
ModelSmall Protein (< 200 aa)Medium (200-500 aa)Large (> 500 aa)
ESMFold30 seconds1-2 minutes2-5 minutes
AlphaFold25 minutes10-15 minutes20-30 minutes
RoseTTAFold10 minutes15-20 minutes30-45 minutes

​
Example: Predicting a Kinase Structure

Let’s predict the structure of CDK2 (Cyclin-Dependent Kinase 2):
1

Search for CDK2

Type “CDK2 human” in the search box. Rosalind finds UniProt entry P24941.
2

Review Sequence

CDK2 is 298 amino acids. AlphaFold2 is perfect for this size.
3

Run Prediction

Click “Predict Structure” with default settings. Takes ~8 minutes.
4

Analyze Results

  • Overall pLDDT: 92.3 (high confidence)
  • ATP binding pocket clearly defined
  • Activation loop shows some flexibility (pLDDT ~75)
5

Next Steps

  • Use structure for kinase inhibitor docking
  • Compare to crystal structure (3PXF) for validation
  • Model drug-resistant mutations (T160M)

​
Validation and Experimental Comparison

Compare your predictions to experimental structures:
  • RMSD calculation: Measure structural similarity
  • TM-score: Topology-independent structure comparison
  • Residue-level alignment: Identify differences
  • Pocket comparison: Validate binding site geometry
LiteFold provides automatic validation when crystal structures are available.

​
Community Models

Access community-contributed models:
  • Domain-specific models (e.g., GPCRs, membrane proteins)
  • Organism-specific models (e.g., viral proteins)
  • Custom-trained models from your team

​
Next Steps

Molecular Docking

Dock ligands to your predicted structure

Molecular Dynamics

Validate and refine with MD simulations

De Novo Design

Design molecules for your structure

Workflows

End-to-end drug discovery workflows