GraphSol

GraphSol predicts protein solubility from sequence with a structure-aware graph convolutional network built on predicted contact maps. Reach for it when a solubility call gates the next step — ranking expression constructs, assessing the impact of mutations, or prioritizing targets for structural studies. It predicts solubility by a different method than DeepSol, so run both and treat agreement as a consensus. See what you can achieve for the full set.

GraphSol integration in .bloom* platform

GraphSol is a structure-aware protein solubility prediction method using attentive Graph Convolutional Networks (GCN). Published in the Journal of Cheminformatics, it was the first study to make sequence-based solubility predictions through graph neural networks, constructing protein topology graphs from predicted contact maps.

Citations

Chen J, Zheng S, Zhao H, Yang Y. Structure-aware protein solubility prediction from sequence through graph convolutional network and predicted contact map. Journal of Cheminformatics. 2021 Feb 8;13:7. doi: 10.1186/s13321-021-00488-1

Performance benchmarks

Tested on the eSOL database, GraphSol achieved state-of-the-art results:

• AUC: 0.866
• F-measure: 0.732
• Accuracy: 0.779 (using threshold of 0.5 for soluble proteins)

Through .bloom*'s webapp, you can rapidly assess protein expression potential and guide protein engineering decisions without requiring experimental structure determination.

Webapp sequence processing capabilities

Input methods in .bloom*

Direct sequence entry:

• Paste amino acid sequences in FASTA or raw format
• Real-time validation with sequence length and composition checks
• Automatic detection of standard vs. non-standard amino acids
• Preview of sequence properties (length, molecular weight, isoelectric point)

Batch processing:

• Upload FASTA files with multiple protein sequences
• Support for multi-sequence files with automatic parsing
• Progress tracking for large sequence libraries
• Error handling for individual sequence failures

Integration with other tools:

• Direct import from BLAST search results
• Connection to sequence clustering outputs
• Import from protein engineering workflows
• Integration with homology modeling results

Prediction outputs

Solubility assessment:

• Probability scores: Continuous solubility prediction (0-1 scale)
• Binary classification: Soluble vs. insoluble predictions
• Confidence estimates: Model certainty for each prediction
• Sequence features: Contributing factors to solubility prediction

Analysis capabilities:

• Batch statistics: Distribution analysis across sequence libraries
• Comparative visualization: Solubility trends across protein families
• Filtering tools: Sort and prioritize sequences by solubility
• Export options: CSV, JSON, and interactive visualization formats

When to use GraphSol in protein engineering

Expression construct design

Solubility screening:

• Evaluate protein variants before experimental expression
• Prioritize constructs with higher solubility potential
• Guide codon optimization and expression host selection
• Reduce experimental screening burden

Protein engineering:

• Assess solubility impact of mutations and insertions
• Compare solubility across protein variant libraries
• Guide directed evolution strategies
• Optimize protein stability and expression

Homolog selection:

• Rank homologous proteins by predicted solubility
• Choose optimal starting points for engineering campaigns
• Compare solubility across species variants
• Guide phylogenetic selection strategies

Structural biology workflows

Target selection:

• Identify soluble protein targets for structural studies
• Prioritize proteins for crystallization trials
• Guide NMR sample preparation decisions
• Optimize protein production strategies

Domain engineering:

• Predict solubility of individual protein domains
• Design soluble fusion constructs
• Optimize linker sequences for multi-domain proteins
• Guide protein fragmentation strategies

Model characteristics and performance

Prediction accuracy

Performance metrics:

• Correlation: Good agreement with experimental solubility data
• Classification accuracy: >80% for soluble/insoluble discrimination
• Sequence coverage: Works across diverse protein families
• Length range: Optimized for 20-4,096 residue proteins

Applicability domain

Protein types supported:

• Single domains: Excellent performance on individual protein domains
• Multi-domain proteins: Good performance with domain boundary consideration
• Fusion proteins: Handles common expression tags and linkers
• Mutant libraries: Sensitive to solubility-changing mutations

Limitations:

• Very short peptides: Reduced accuracy for sequences <20 residues
• Membrane proteins: Limited applicability to transmembrane regions
• Intrinsically disordered: May not capture IDP solubility characteristics
• Post-translational modifications: Does not account for glycosylation effects

Performance and scaling

Processing efficiency

Speed characteristics:

• Single sequences: Near-instant predictions (<1 second)
• Batch processing: Linear scaling with sequence count
• Large libraries: Support for thousands of sequences per job
• Real-time feedback: Immediate results for interactive analysis

Resource optimization

Computational requirements:

• Low computational overhead per prediction
• No GPU requirements for standard predictions
• Memory efficient for large sequence batches
• Optimized for high-throughput screening

Integration with .bloom* ecosystem

Upstream sequence sources

Bioinformatics tools:

• BLAST: Evaluate homolog solubility from search results
• Sequence alignment: Compare solubility across aligned sequences
• Motif analysis: Assess functional motif impact on solubility
• Domain prediction: Evaluate domain solubility contributions

Protein design:

• ESM prediction integration: Combine folding and solubility assessment
• Mutagenesis planning: Guide mutation strategies for improved solubility
• Library design: Optimize variant libraries for soluble expression

Downstream applications

Experimental planning:

• Guide protein expression condition optimization
• Plan purification strategy based on solubility predictions
• Design solubility screening assays
• Optimize buffer and additive selection

Protein production:

• Select optimal expression constructs
• Guide scale-up production decisions
• Plan formulation development strategies
• Optimize protein storage conditions

Workflow automation

High-throughput screening:

• Automated batch processing of sequence libraries
• Integration with laboratory automation systems
• Real-time result monitoring and prioritization
• Quality control and data validation pipelines

Research collaboration:

• Shared solubility predictions across teams
• Integration with electronic lab notebooks
• Audit trails for experimental reproducibility
• Comparative analysis across research projects

Chat Usage

Example Prompts

• "Predict solubility for human lysozyme using GraphSol"
• "Run GraphSol on my ESMFold structure to find aggregation-prone regions"
• "Compare GraphSol vs DeepSol predictions for this protein"

Workflow Chaining

• Feeds into: Formulation planning, Compound Passport
• Accepts from: UniProt (sequences), ESMFold (predicted structures)

FAQ

Read next