> ## Documentation Index
> Fetch the complete documentation index at: https://docs.litefold.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Protein Structure Prediction
> Predict 3D protein structures with AI-powered models
Protein structure prediction enables researchers to infer the three-dimensional arrangement of atoms in a protein directly from its amino acid sequence. By predicting how a protein folds, researchers can gain valuable insights into its function, interactions, and role in biological processes.
Understanding protein structure is essential for a wide range of applications, including studying protein function, identifying potential binding sites, analyzing the effects of mutations, engineering novel proteins, and supporting drug discovery efforts. Structural information can help researchers generate hypotheses, prioritize experiments, and better interpret biological data.
LiteFold provides a streamlined platform for generating protein structure predictions, helping researchers quickly move from sequence data to structural insights. Predicted structures can be used to support downstream analyses such as molecular docking, protein design, functional annotation, and other computational biology workflows.
## Key Features
* AI-powered protein structure prediction
* Simple sequence-based prediction workflow
* Interactive structure visualization
* Downloadable structure files for downstream analysis
* Integration with LiteFold molecular docking workflows
* Integration with molecular dynamics simulations
* Fast prediction turnaround for research applications
#### Available Models
Boltz-2
# Supported Prediction Tasks
LiteFold supports both individual protein structure prediction and multi-component biological assemblies.
Supported prediction workflows include:
| Prediction Type | Description |
| -------------------------- | ----------------------------------------------------------- |
| Single Protein | Predict the structure of an individual protein chain |
| Protein-Protein Complex | Predict interactions between multiple protein chains |
| Protein-Ligand Complex | Predict protein structures together with bound ligands |
| Protein-DNA Complex | Model protein-DNA interactions |
| Protein-RNA Complex | Model protein-RNA interactions |
| Multi-Component Assemblies | Predict systems containing multiple biomolecular components |
These prediction modes enable researchers to study binding interfaces, signaling complexes, regulatory proteins, enzyme-substrate interactions, and drug-target systems from a unified workflow.
## **Quick Start**
This guide walks through the complete workflow for generating protein structures in LiteFold using Boltz-2.
## Before You Begin
Before starting a prediction, ensure that:
* A project has been created.
* Your protein sequence has been uploaded as a FASTA file.
* You have sufficient compute credits available.
> Need help uploading files?
>
> See: **Uploading Files to LiteFold**
## Step 1: Open the Structure Prediction Module
From the LiteFold Lab dashboard:
1. Navigate to **Lab**.
2. Select **Structure Prediction** from the sidebar.
The Structure Prediction dashboard displays all previous prediction jobs associated with your projects.
## Step 2: Select a Project
Use the **Project** dropdown menu to choose the project where your prediction will be created.
Projects help organize prediction jobs, input files, and generated structures.
## Step 3: Create a New Structure Prediction
After selecting a project:
1. Open the project.
2. Create a new structure prediction job.
LiteFold will load all FASTA files associated with the selected project.
LiteFold supports both single-chain and multi-chain prediction workflows. A prediction may contain one or more FASTA entries, allowing users to model biological complexes in addition to individual proteins.
# Step 4: Review Input Sequences
Select one or more FASTA files that will be included in the prediction.
LiteFold displays:
* Protein sequences
* Chain assignments
* Residue counts
* Sequence composition
* Molecular component types
When multiple sequences are present, each sequence is assigned as a separate chain or biological component within the prediction workflow.
This allows users to verify all submitted components before launching a prediction.
# Multi-Sequence and Complex Prediction
LiteFold supports the submission of multiple FASTA entries within a single prediction job.
This capability enables prediction of:
* Protein-protein complexes
* Protein-ligand systems
* Protein-RNA assemblies
* Protein-DNA assemblies
* Multi-component biological complexes
Multiple molecular components can be combined into a single prediction, allowing LiteFold to predict both individual structures and their spatial arrangement within the final assembly.
### Example
```
>Protein_A
MSEQNNTEMTFQIQRIYTKDISFEAPNAPHVFQKDWLDKKTGKEVVEFDNQKDVKAIANSLGKD
>Protein_B
MNKLLILTCLVAVALARPKTNAEAKKAAEAAKAAEAAKAAEAAKAAEAAKAAEA
```
LiteFold automatically identifies each component and predicts the complete complex structure.
## Step 5: Review Boltz Configuration (Optional)
Before running the prediction, you may review the **Boltz Configuration** tab.
The configuration panel provides information about:
* Selected prediction engine
* Sequence interpretation
* Chain assignments
* Model-specific settings
Most users can proceed with the default configuration.
## Step 6: Start a Prediction
Click **Start Prediction** in the upper-right corner.
A prediction setup dialog will appear.
## Step 7: Select the Prediction Model
Choose the model that will be used for structure generation.
Available models include:
### Boltz-2
Recommended for most prediction tasks.
Provides:
* High-quality structure prediction
* Multi-chain support
* Fast inference
* Strong performance on protein complexes
## Step 8: Select Input Files
Choose the FASTA file(s) to include in the prediction job.
You may:
* Select individual files
* Select multiple files
* Use **Select All**
Only validated FASTA files can be submitted for prediction.
## Step 9: Validate Files
Click **Validate Files**.
LiteFold performs automatic validation checks, including:
* FASTA format verification
* Sequence integrity checks
* Input compatibility checks
* Model readiness verification
Once validation succeeds, the prediction can be submitted.
## Step 10: Launch the Prediction
Click **Start Prediction** to submit the job.
LiteFold will:
1. Queue the prediction task.
2. Allocate computational resources.
3. Begin structure generation.
4. Track job progress automatically.
A confirmation message will indicate that the job has been submitted successfully.
## Monitoring Prediction Progress
After submission, return to the **Structure Prediction Home Page**.
The dashboard displays:
* Running predictions
* Queued jobs
* Completed predictions
* Failed jobs (if any)
* Historical prediction records
From this page you can monitor progress, review completed structures, and access previously generated prediction results.
# Understanding Results
Once a prediction has completed, LiteFold automatically generates a detailed results page where you can inspect the predicted structure, evaluate confidence metrics, save results to your project, and download structure files for downstream analysis.
The results workspace combines interactive visualization with quantitative quality metrics, allowing researchers to quickly assess whether a predicted structure is suitable for further study.
# Overview
The results page provides:
* Interactive 3D structure visualization
* Confidence metrics
* Structure quality assessment
* Prediction model information
* File export options
* Project organization tools
The left panel displays completed prediction jobs, while the main workspace provides detailed information about the selected structure.
# Structure Visualization
The **Structure** tab displays an interactive three-dimensional representation of the predicted protein.
The viewer allows you to:
* Rotate the structure
* Zoom in and out
* Inspect individual regions
* Examine chain organization
* Analyze folded domains
* Review confidence coloring
This provides a quick visual assessment of structural quality and folding patterns.
# Confidence Coloring (pLDDT)
LiteFold automatically colors the predicted structure according to pLDDT confidence scores.
For residue i:
$$
pLDDTi∈[0,100]
$$
Based on the Local Distance Difference Test:
$$
LDDT_i=
\frac{1}{N_i}
\sum_{j}
I\left(
|d_{ij}^{pred}-d_{ij}^{true}|<\tau
\right)
$$
**Where:**
* $d_{ij}^{pred}$ denotes the predicted distance.
* $d_{ij}^{true}$ denotes the experimental distance.
* $\tau$ represents the distance threshold.
* $N_i$ represents the neighboring residues.
* $I$ is the indicator function.
The color legend is displayed above the structure viewer.
| Color | pLDDT Range | Interpretation |
| ---------- | ----------- | -------------------- |
| Dark Blue | 90+ | Very high confidence |
| Light Blue | 70–90 | Reliable prediction |
| Yellow | 50–70 | Moderate confidence |
| Orange | \<50 | Low confidence |
Regions displayed in dark blue are generally the most reliable portions of the prediction.
# Prediction Metrics
Below the structure viewer, LiteFold provides several quality metrics to help evaluate prediction reliability.
## Mean pLDDT
**Mean pLDDT** represents the average confidence score across all residues in the structure.
$$
\text{Mean pLDDT} = \frac{1}{N}\sum_{i=1}^{N} pLDDT_i
$$
Where:
N = total residues
Example:
`Mean pLDDT: 93.3%`
Interpretation:
* Above 90 → Very high confidence
* 70–90 → Reliable prediction
* 50–70 → Moderate confidence
* Below 50 → Low confidence
Higher values generally indicate more reliable structural predictions.
## Confidence Score
The overall confidence score summarizes the model's confidence in the generated structure.
$$
Confidence=w1(Mean pLDDT)+w2(pTM)+w3(ipTM)
$$
Where:
w1+w2+w3=1
w1+w2+w3=1
Example:
`Confidence: 90.2%`
Structures with higher confidence scores are typically better candidates for downstream applications such as docking and molecular dynamics.
## Clash Score
The Clash Score measures steric overlaps between atoms.
Example:
`Clash Score: 3.42`
Lower clash scores generally indicate better structural geometry and fewer unrealistic atomic contacts.
## pTM Score
The **Predicted Template Modeling (pTM)** score evaluates overall structural accuracy.
Example:
`pTM: 0.802`
pTM values range from 0 to 1.
| Score | Interpretation |
| ------- | -------------------------- |
| > 0.8 | High structural confidence |
| 0.6–0.8 | Moderate confidence |
| \< 0.6 | Lower confidence |
## ipTM Score
For multi-chain complexes, LiteFold reports the **Interface Predicted Template Modeling (ipTM)** score.
Example:
`ipTM: 0.779`
This metric estimates confidence in the predicted interactions between biological components within a complex.
Depending on the prediction setup, ipTM may be used to assess:
* Protein-protein interfaces
* Protein-ligand interactions
* Protein-RNA interfaces
* Protein-DNA interfaces
Higher ipTM values generally indicate greater confidence in the predicted intermolecular contacts.
Higher ipTM values generally indicate more reliable protein-protein interfaces.
For multi-component predictions, LiteFold evaluates both individual chain quality and interface quality.
Additional metrics may include:
| Metric | Purpose |
| ----------- | ----------------------------------- |
| pTM | Overall structural confidence |
| ipTM | Interface confidence between chains |
| Mean pLDDT | Average residue confidence |
| Clash Score | Structural geometry quality |
For protein-protein, protein-ligand, protein-RNA, and protein-DNA predictions, interface-focused metrics are particularly important when assessing biological relevance.
## Prediction Model
The metrics table also records the model used to generate the structure.
Example:
`Model: Boltz`
This allows researchers to track which prediction engine produced a particular result.
# Saving Results
LiteFold allows completed predictions to be assigned a custom alias and stored within your project for easier organization.
Click **Save As** to open the save dialog.
## Creating an Alias
An alias provides a more meaningful name for the prediction.
Examples:
* KRAS\_G12D\_Prediction
* EGFR\_T790M\_Model
* B2AR\_Active\_State
Aliases must be unique within the selected project.
## Adding Results to Libraries
Predicted structures can optionally be added to a LiteFold Library for later reuse across projects and workflows.
This is useful for:
* Frequently used targets
* Reference structures
* Shared project resources
## Converting to Additional Formats
During the save process, LiteFold can optionally convert the prediction into additional structure formats.
Supported formats may include:
* CIF
* PDB
* PDBQT
This simplifies preparation for docking and simulation workflows.
# Downloading Structure Files
Predicted structures can be downloaded directly from the results page.
Click **Download .cif** to export the structure.
## CIF Format
The Crystallographic Information File (CIF) format contains:
* Atomic coordinates
* Structural annotations
* Chain information
* Residue information
The CIF format is widely supported by structural biology software.
### Best Practices
Before proceeding to downstream analysis:
* Review the Mean pLDDT score.
* Check confidence coloring in critical regions.
* Examine clash scores for structural quality.
* Verify interface quality using ipTM for complexes.
* Save important predictions using descriptive aliases.
* Download structures for archival and external analysis.
After reviewing the prediction results, you can:
1. Save the structure to your project.
2. Add it to a LiteFold Library.
3. Download the CIF file.
4. Use the structure in Molecular Docking workflows.
5. Run Molecular Dynamics simulations.
6. Continue with De Novo Design and drug discovery pipelines.
These results provide the foundation for structure-guided biological research and computational drug discovery within LiteFold.
## Next Steps
Dock ligands to your predicted structure
Validate and refine with MD simulations
Design molecules for your structure
End-to-end drug discovery workflows
## Step 5: Review Boltz Configuration (Optional)
Before running the prediction, you may review the **Boltz Configuration** tab.
The configuration panel provides information about:
* Selected prediction engine
* Sequence interpretation
* Chain assignments
* Model-specific settings
Most users can proceed with the default configuration.
## Step 6: Start a Prediction
Click **Start Prediction** in the upper-right corner.
A prediction setup dialog will appear.
## Step 7: Select the Prediction Model
Choose the model that will be used for structure generation.
Available models include:
### Boltz-2
Recommended for most prediction tasks.
Provides:
* High-quality structure prediction
* Multi-chain support
* Fast inference
* Strong performance on protein complexes
## Step 8: Select Input Files
Choose the FASTA file(s) to include in the prediction job.
You may:
* Select individual files
* Select multiple files
* Use **Select All**
Only validated FASTA files can be submitted for prediction.
## Step 9: Validate Files
Click **Validate Files**.
LiteFold performs automatic validation checks, including:
* FASTA format verification
* Sequence integrity checks
* Input compatibility checks
* Model readiness verification
Once validation succeeds, the prediction can be submitted.
## Step 10: Launch the Prediction
Click **Start Prediction** to submit the job.
LiteFold will:
1. Queue the prediction task.
2. Allocate computational resources.
3. Begin structure generation.
4. Track job progress automatically.
A confirmation message will indicate that the job has been submitted successfully.
## Monitoring Prediction Progress
After submission, return to the **Structure Prediction Home Page**.
The dashboard displays:
* Running predictions
* Queued jobs
* Completed predictions
* Failed jobs (if any)
* Historical prediction records
From this page you can monitor progress, review completed structures, and access previously generated prediction results.
# Understanding Results
Once a prediction has completed, LiteFold automatically generates a detailed results page where you can inspect the predicted structure, evaluate confidence metrics, save results to your project, and download structure files for downstream analysis.
The results workspace combines interactive visualization with quantitative quality metrics, allowing researchers to quickly assess whether a predicted structure is suitable for further study.
# Overview
The results page provides:
* Interactive 3D structure visualization
* Confidence metrics
* Structure quality assessment
* Prediction model information
* File export options
* Project organization tools
The left panel displays completed prediction jobs, while the main workspace provides detailed information about the selected structure.
# Structure Visualization
The **Structure** tab displays an interactive three-dimensional representation of the predicted protein.
The viewer allows you to:
* Rotate the structure
* Zoom in and out
* Inspect individual regions
* Examine chain organization
* Analyze folded domains
* Review confidence coloring
This provides a quick visual assessment of structural quality and folding patterns.
# Confidence Coloring (pLDDT)
LiteFold automatically colors the predicted structure according to pLDDT confidence scores.
For residue i:
$$
pLDDTi∈[0,100]
$$
Based on the Local Distance Difference Test:
$$
LDDT_i=
\frac{1}{N_i}
\sum_{j}
I\left(
|d_{ij}^{pred}-d_{ij}^{true}|<\tau
\right)
$$
**Where:**
* $d_{ij}^{pred}$ denotes the predicted distance.
* $d_{ij}^{true}$ denotes the experimental distance.
* $\tau$ represents the distance threshold.
* $N_i$ represents the neighboring residues.
* $I$ is the indicator function.
The color legend is displayed above the structure viewer.
| Color | pLDDT Range | Interpretation |
| ---------- | ----------- | -------------------- |
| Dark Blue | 90+ | Very high confidence |
| Light Blue | 70–90 | Reliable prediction |
| Yellow | 50–70 | Moderate confidence |
| Orange | \<50 | Low confidence |
Regions displayed in dark blue are generally the most reliable portions of the prediction.
# Prediction Metrics
Below the structure viewer, LiteFold provides several quality metrics to help evaluate prediction reliability.
## Mean pLDDT
**Mean pLDDT** represents the average confidence score across all residues in the structure.
$$
\text{Mean pLDDT} = \frac{1}{N}\sum_{i=1}^{N} pLDDT_i
$$
Where:
N = total residues
Example:
`Mean pLDDT: 93.3%`
Interpretation:
* Above 90 → Very high confidence
* 70–90 → Reliable prediction
* 50–70 → Moderate confidence
* Below 50 → Low confidence
Higher values generally indicate more reliable structural predictions.
## Confidence Score
The overall confidence score summarizes the model's confidence in the generated structure.
$$
Confidence=w1(Mean pLDDT)+w2(pTM)+w3(ipTM)
$$
Where:
w1+w2+w3=1
w1+w2+w3=1
Example:
`Confidence: 90.2%`
Structures with higher confidence scores are typically better candidates for downstream applications such as docking and molecular dynamics.
## Clash Score
The Clash Score measures steric overlaps between atoms.
Example:
`Clash Score: 3.42`
Lower clash scores generally indicate better structural geometry and fewer unrealistic atomic contacts.
## pTM Score
The **Predicted Template Modeling (pTM)** score evaluates overall structural accuracy.
Example:
`pTM: 0.802`
pTM values range from 0 to 1.
| Score | Interpretation |
| ------- | -------------------------- |
| > 0.8 | High structural confidence |
| 0.6–0.8 | Moderate confidence |
| \< 0.6 | Lower confidence |
## ipTM Score
For multi-chain complexes, LiteFold reports the **Interface Predicted Template Modeling (ipTM)** score.
Example:
`ipTM: 0.779`
This metric estimates confidence in the predicted interactions between biological components within a complex.
Depending on the prediction setup, ipTM may be used to assess:
* Protein-protein interfaces
* Protein-ligand interactions
* Protein-RNA interfaces
* Protein-DNA interfaces
Higher ipTM values generally indicate greater confidence in the predicted intermolecular contacts.
Higher ipTM values generally indicate more reliable protein-protein interfaces.
For multi-component predictions, LiteFold evaluates both individual chain quality and interface quality.
Additional metrics may include:
| Metric | Purpose |
| ----------- | ----------------------------------- |
| pTM | Overall structural confidence |
| ipTM | Interface confidence between chains |
| Mean pLDDT | Average residue confidence |
| Clash Score | Structural geometry quality |
For protein-protein, protein-ligand, protein-RNA, and protein-DNA predictions, interface-focused metrics are particularly important when assessing biological relevance.
## Prediction Model
The metrics table also records the model used to generate the structure.
Example:
`Model: Boltz`
This allows researchers to track which prediction engine produced a particular result.
# Saving Results
LiteFold allows completed predictions to be assigned a custom alias and stored within your project for easier organization.
Click **Save As** to open the save dialog.
## Creating an Alias
An alias provides a more meaningful name for the prediction.
Examples:
* KRAS\_G12D\_Prediction
* EGFR\_T790M\_Model
* B2AR\_Active\_State
Aliases must be unique within the selected project.
## Adding Results to Libraries
Predicted structures can optionally be added to a LiteFold Library for later reuse across projects and workflows.
This is useful for:
* Frequently used targets
* Reference structures
* Shared project resources
## Converting to Additional Formats
During the save process, LiteFold can optionally convert the prediction into additional structure formats.
Supported formats may include:
* CIF
* PDB
* PDBQT
This simplifies preparation for docking and simulation workflows.
# Downloading Structure Files
Predicted structures can be downloaded directly from the results page.
Click **Download .cif** to export the structure.
## CIF Format
The Crystallographic Information File (CIF) format contains:
* Atomic coordinates
* Structural annotations
* Chain information
* Residue information
The CIF format is widely supported by structural biology software.
### Best Practices
Before proceeding to downstream analysis:
* Review the Mean pLDDT score.
* Check confidence coloring in critical regions.
* Examine clash scores for structural quality.
* Verify interface quality using ipTM for complexes.
* Save important predictions using descriptive aliases.
* Download structures for archival and external analysis.
After reviewing the prediction results, you can:
1. Save the structure to your project.
2. Add it to a LiteFold Library.
3. Download the CIF file.
4. Use the structure in Molecular Docking workflows.
5. Run Molecular Dynamics simulations.
6. Continue with De Novo Design and drug discovery pipelines.
These results provide the foundation for structure-guided biological research and computational drug discovery within LiteFold.
## Next Steps