PASSION

PASSION predicts binding sites between circular RNAs (circRNAs) and RNA-binding proteins (RBPs) across 37 RBP types using an ensemble neural network approach for circRNA–RBP interaction mapping.

Key Features:

• Target scope: Predicts circRNA–RBP binding sites across 37 distinct RBPs.
• Architecture: Uses an ensemble neural network combining a concatenated artificial neural network (ANN) and a hybrid deep neural network.
• ANN feature strategy: The ANN component utilizes an optimal feature subset for each RBP derived from six distinct feature encoding schemes.
• Feature selection and optimization: Features are selected via incremental feature selection and optimized using the XGBoost algorithm.
• Hybrid DNN input: The hybrid deep neural network component uses a stacked codon-based encoding scheme as input.
• Benchmarking: Performance was compared against XGBoost, k-nearest neighbor, support vector machine, random forest, logistic regression, and Naive Bayes.
• Overall performance: The ensemble architecture achieves an average area under the curve (AUC) of 0.883 across datasets for all 37 RBPs.
• Independent-test robustness: Independent tests at sequence similarity thresholds 0.8, 0.7, 0.6, and 0.5 yielded AUCs of 0.883, 0.876, 0.868, and 0.883 respectively.
• CircRNA context: Accounts for the closed ring structure of circRNAs that enables direct interactions with proteins.

Scientific Applications:

• CircRNA–RBP binding site prediction: Identification of nucleotide-level binding sites between circRNAs and RBPs.
• Large-scale interaction mapping: Systematic mapping of circRNA–RBP interactions across a panel of 37 RBPs.
• Robustness assessment: Evaluation of prediction stability under varying sequence similarity thresholds.

Methodology:

Integrates a concatenated ANN and a hybrid deep neural network; the ANN uses optimal feature subsets per RBP derived from six feature encoding schemes selected via incremental feature selection and optimized with XGBoost, while the hybrid DNN accepts a stacked codon-based input scheme.