circ-pSBLA

circ-pSBLA predicts binding sites between circular RNAs (circRNAs) and RNA-binding proteins (RBPs) using a pseudo-Siamese framework that integrates BiLSTM, a Soft Attention Mechanism, and dual classifiers (Softmax and CatBoost) to improve prediction accuracy.

Key Features:

• Pseudo-Siamese Framework: Integrates paired network branches to jointly model circRNA and RBP sequence representations for interaction prediction.
• BiLSTM Network: Uses a Bi-directional Long Short-Term Memory (BiLSTM) network to capture sequential dependencies and extract sequence features.
• Soft Attention Mechanism: Applies a Soft Attention Mechanism to weight and emphasize relevant regions of input sequences during feature learning.
• Dual Classification Approach: Combines Softmax and CatBoost classifiers to perform final binding-site prediction leveraging complementary classification strategies.
• Comprehensive Validation: Evaluated by comparisons with other methods across 17 sub-datasets and includes motif analysis on three sub-datasets.

Scientific Applications:

• Disease Research: Supports investigation of molecular mechanisms in diseases by predicting circRNA-RBP interactions relevant to pathology.
• Functional Genomics: Facilitates study of circRNA regulatory roles and RBP-associated regulatory networks through predicted binding sites.

Methodology:

circ-pSBLA integrates BiLSTM for feature extraction and a Soft Attention Mechanism within a pseudo-Siamese framework and employs Softmax and CatBoost classifiers, with performance validated across 17 sub-datasets and by motif analysis on three sub-datasets.