Periscope

Periscope predicts real-valued expression levels of soluble proteins in the periplasm of Escherichia coli to support selection of constructs and signal peptide–target combinations for periplasmic expression studies.

Key Features:

• Two-Stage Architecture: A support vector machine (SVM) classifier first assigns inputs to a specific second-stage model, after which support vector regression (SVR) produces real-valued expression predictions.
• Real-Valued Expression Prediction: Estimates continuous periplasmic soluble protein expression levels rather than binary outcomes.
• Performance Metrics: Reported overall prediction accuracy of 78% and a Pearson's correlation coefficient (PCC) of 0.77 on an independent test dataset.
• Feature Importance Analysis: Identifies the relative importance of sequence-derived features, notably the occurrence of dipeptides containing glutamine and aspartic acid, for classification and prediction.

Scientific Applications:

• Experimental prioritization: Complements trial-and-error screening by prioritizing candidate constructs for periplasmic expression in Escherichia coli.
• Construct selection: Assists selection of promising target proteins and signal peptide combinations to increase the likelihood of obtaining soluble, periplasmic protein.
• Proteomics and production planning: Guides rational decision-making to improve efficiency and reduce resource expenditure in protein expression experiments.

Methodology:

Periscope applies a two-stage predictive architecture using an SVM classifier to select between support vector regression models for real-valued periplasmic expression prediction, performs feature-importance analysis (highlighting Q–D dipeptide occurrence), and was evaluated on an independent test dataset yielding 78% accuracy and PCC 0.77.