Limitations

While GDAP provides valuable predictions, it has several important limitations that users should be aware of.

Data Limitations

Coverage

• Disease-specific: Predictions are limited to diseases with sufficient data in Open Targets Platform
• Gene coverage: Only includes genes with protein interaction data in STRING database
• Species-specific: Currently limited to Homo sapiens

Data Quality

• Experimental bias: Training data reflects historical research priorities
• Publication bias: Well-studied diseases have more data
• Annotation quality: Depends on accuracy of source databases

Model Limitations

Prediction Accuracy

• False positives: Some predicted associations may not be biologically relevant
• False negatives: May miss associations due to data limitations
• Confidence scores: Should not be interpreted as absolute probabilities

Generalization

• Training bias: Models trained on known associations may not generalize to novel relationships
• Disease specificity: Models trained on one disease may not apply to others
• Temporal changes: New discoveries may not be reflected in training data

Technical Limitations

Computational Resources

• Memory requirements: Large datasets require significant RAM
• Processing time: Complex embeddings can take hours to generate
• Storage: Results can be several GB for large analyses

Scalability

• Graph size: Very large networks may exceed memory limits
• Embedding quality: Performance may degrade with extremely large graphs
• Model complexity: Some algorithms don’t scale well to massive datasets

Biological Limitations

Context Dependence

• Tissue specificity: Predictions don’t account for tissue-specific expression
• Temporal dynamics: Don’t capture developmental or disease progression changes
• Environmental factors: Don’t consider external influences on gene expression

Validation Requirements

• Experimental validation: All predictions require laboratory confirmation
• Biological context: Predictions need interpretation in disease context
• Clinical relevance: Not all associations are clinically meaningful

Best Practices

Mitigating Limitations

1. Use multiple models: Compare predictions across different algorithms
2. Validate predictions: Always confirm with experimental data
3. Consider context: Interpret results in biological and clinical context
4. Update regularly: Retrain models with new data

Quality Control

• Check data sources: Verify disease and gene data quality
• Monitor performance: Track model accuracy over time
• Document assumptions: Clearly state limitations in reports