DeepSol quickstart
Fast path to predict protein solubility for E. coli expression using CNN-based deep learning.
You want to know whether a protein will express solubly in E. coli before you commit to an expression trial. This walks you from a single amino acid sequence (or a FASTA batch) to a solubility prediction in five steps: open the tool, add your sequences, configure settings, submit, and interpret the scores.
Quick protein solubility prediction with .bloom* platform
DeepSol uses convolutional neural networks (CNNs) to predict protein solubility when expressed in E. coli. Published in Bioinformatics 2018, the model exploits k-mer structure and combines sequence-level and structural features to achieve 77% accuracy.
Step-by-step solubility analysis
Access the DeepSol tool
Navigate to the Protein Solubility Prediction tool in .bloom*'s interactive workspace. The tool provides guided input validation and instant result visualization powered by the DeepSol CNN architecture.
Prepare your protein sequences
Option A: Single sequence (fastest)
- Click Sequence Input in the proteins section
- Enter your amino acid sequence in single-letter code (e.g., "MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH...")
- .bloom* automatically validates the sequence and shows basic properties
- Preview sequence length and composition before submission
Option B: Batch processing
- Click File Upload in the proteins section
- Upload a FASTA file with multiple protein sequences
- Optional: Include descriptive headers for tracking
- .bloom* parses the file and shows sequence count and validation status
Option C: From previous protein jobs
- Click From Previous Job in the proteins section
- Select outputs from AlphaFold, protein design tools, or other workflows
- Choose specific sequences or use all generated proteins
- .bloom* handles format conversion automatically
Configure prediction settings
Basic settings (recommended):
- Keep default parameters for standard predictions
- Job name: Enter descriptive name or leave blank for auto-generation
Advanced settings (via gear icon):
- Batch size: Adjust for optimal processing (default 32)
- Binary classification: Return soluble/insoluble labels
- Probability scores: Include raw CNN output probabilities
- Feature extraction: Enable to output learned representations
Submit and monitor the job
- Click Predict Solubility to start analysis
- .bloom* shows real-time progress as sequences pass through the CNN layers
- Typical completion: seconds to minutes depending on batch size
- Receive notifications when complete (if enabled in settings)
Analyze and interpret results
Interactive result exploration:
- View results in sortable table format
- Filter by solubility prediction, confidence scores, or warnings
- Visualize distribution with histograms and scatter plots
- Compare proteins side-by-side with property overlays
Result interpretation:
- Solubility score: Probability of soluble expression (0-1 scale)
- Binary prediction: Soluble or insoluble classification
- Confidence scores: Model certainty based on CNN output
- Warnings: Flags for sequences outside training domain
Export options:
- CSV download: Full results with all properties
- Filtered datasets: Export only sequences meeting criteria
- Visualization exports: Charts and plots for reports
Common prediction scenarios
Protein library screening
- Upload FASTA file with protein variants
- Run batch prediction with default settings
- Filter results by solubility score (focus on high probability soluble)
- Export prioritized sequence list for expression trials
Protein engineering optimization
- Prepare sequences for protein variants
- Submit batch prediction
- Sort by solubility scores to identify trends
- Visualize sequence-solubility relationships
- Guide selection of best candidates for E. coli expression
Expression construct assessment
- Input candidate protein sequences from design pipelines
- Check solubility predictions against expression requirements
- Flag sequences likely to form inclusion bodies
- Generate reports for downstream expression planning
Quality control and validation
Input preparation checklist
Sequence format:
- Valid amino acid sequence (standard 20 amino acids)
- No ambiguous residues (X, B, Z, J) unless specifically supported
- Single sequence per entry (no multi-chain complexes)
- Sequence length within model limits
Sequence properties:
- Sequence length in optimal range (typically 50-2000 residues)
- No non-standard amino acids or post-translational modifications
- Represents mature protein (signal peptides removed if applicable)
- Consistent with E. coli codon usage considerations
Result validation
Prediction reliability:
- Check confidence scores (higher = more reliable)
- Review warning flags and their implications
- Validate against known soluble/insoluble proteins when possible
- Consider experimental verification for critical decisions
Batch consistency:
- Ensure consistent sequence formatting across proteins
- Check for duplicates that may skew results
- Verify batch size is appropriate for dataset
- Monitor for systematic patterns in predictions
Troubleshooting common issues
Invalid sequence errors
- Check for non-standard amino acid characters
- Remove header lines if pasting raw sequences
- Ensure proper FASTA format for file uploads
- Consider sequence cleanup tools for normalization
Poor predictions
- Verify sequences are within the training domain
- Check sequence length and composition
- Consider if proteins have extreme properties (very hydrophobic, disordered)
- Review training domain limitations for specialized protein classes
Batch processing issues
- Split very large FASTA files into smaller batches
- Ensure consistent sequence formatting across the dataset
- Check for encoding issues in uploaded files
- Validate FASTA format and header specifications
Performance issues
- Reduce batch size for very long sequences
- Consider splitting jobs for very large datasets
- Check file format and remove unnecessary metadata
- Monitor system resources during processing
Next steps after prediction
Protein prioritization:
- Use solubility data to rank expression candidates
- Combine with other biophysical predictions for comprehensive assessment
- Guide protein engineering optimization efforts
Expression planning:
- Identify proteins likely to require refolding from inclusion bodies
- Plan alternative expression strategies for low-solubility candidates
- Design fusion tags or solubility-enhancing mutations based on predictions
Further analysis:
- Integrate with structure prediction tools
- Feed results to protein design workflows for optimization
- Generate comprehensive expression profiling reports
Read next
Was this page helpful?