DSResSol

DSResSol predicts protein solubility from amino acid sequences using deep learning to identify sequence determinants that influence solubility.

Key Features:

• Deep Learning Architecture: DSResSol employs squeeze excitation residual networks integrated with dilated convolutional neural networks to model sequence features.
• Local and Long-Range Interaction Capture: The model captures both local and global interactions of frequently occurring amino acid k-mers important for solubility.
• Input Data: Protein sequences (amino acid sequences) are used as the input for prediction.
• Performance and Accuracy: DSResSol outperforms available sequence-based solubility predictors by at least 5% across two independent test sets, reduces prediction bias for insoluble proteins, and improves soluble protein prediction accuracy by at least 13%.
• Identification of Key Amino Acids: The model identifies glutamic acid and serine as important contributors to protein solubility.

Scientific Applications:

• Protein Function Characterization: Predicted solubility informs analyses of protein functional properties and behavior in experimental conditions.
• Industrial Protein Production: Solubility predictions support optimization of protein expression and yield for pharmaceutical and industrial production workflows.

Methodology:

DSResSol takes protein sequences as input and processes them through a neural network architecture combining squeeze excitation residual networks with dilated convolutional neural networks to capture local and long-range interactions of amino acid k-mers.