AC-Caps

AC-Caps predicts RNA-binding protein (RBP) binding sites on long non-coding RNAs (lncRNAs) using a hybrid deep learning architecture to characterize nucleotide patterns and infer RNA–protein interactions.

Key Features:

• High-order statistical-based encoding: Encodes nucleotide sequences of lncRNAs to capture high-order nucleotide statistics and patterns.
• Joint processing layer (attention + CNN): Integrates an attention mechanism with a convolutional neural network (CNN) to emphasize relevant sequence features.
• Capsule network: Preserves spatial hierarchies between extracted features for refined binding-site prediction.
• Benchmark datasets: Evaluated on 31 independent experimental datasets from 12 lncRNA-binding proteins.
• Performance metrics: Reported average area under the curve (AUC) of 0.967 and average accuracy (ACC) of 92.5%.
• Comparative performance: Demonstrated improvements over HOCCNNLB, iDeepS, and DeepBind with AUC increases up to 0.261 and accuracy increases up to 28.9%.

Scientific Applications:

• Elucidation of regulatory mechanisms: Prediction of RBP binding sites on lncRNAs to study epigenetic and post-transcriptional regulatory mechanisms.
• Disease and physiology research: Analysis of RBP–lncRNA interactions in physiological and pathological processes, including cancer research.
• Large-scale binding-site analysis: Processing and analysis of large-scale RBP binding site data on lncRNA chains.
• Method benchmarking: Comparative evaluation of RBP binding prediction methods using standardized experimental datasets and performance metrics.

Methodology:

Sequence data are encoded using a high-order statistical-based scheme, processed through a joint attention+CNN layer, passed to a capsule network for prediction, and evaluated on 31 independent experimental datasets from 12 lncRNA-binding proteins.