RBPSpot

RBPSpot identifies ribonucleoprotein (RBP) binding sites in RNA sequences by combining BWT/FM-index inexact k-mer search and a deep feed-forward neural network to predict context-dependent RBP binding.

Key Features:

• Ultra-Fast BWT/FM-Indexing: Uses Burrows-Wheeler Transform (BWT) combined with FM-indexing to perform an inexact k-mer spectrum search and identify statistically significant seed sequences as anchors.
• Contextual Evaluation Using DNN: Employs a Deep Feed-forward Neural Network (DNN) to assess the binding potential of identified seeds by considering flanking region information.
• Incorporation of Contextual Features: Integrates pentamers, dinucleotide patterns, and CG distribution patterns to capture sequence context related to shape and structural properties influencing RBP interactions.
• Support from MD-Simulation Studies: Validates models and predictions with molecular dynamics (MD) simulation studies to align predictions with biophysical behavior.
• High Performance and Benchmarking: Benchmarked on over 50 RBPs across three datasets with reported average accuracy of approximately 90% and compared against other tools including complex deep-learning models.
• Extensive Coverage and Model Flexibility: Covers 131 human RBPs for which sufficient CLIP-seq data exist and supports building new models from user-provided data for other species or RBPs.

Scientific Applications:

• RBP binding site discovery: Identification and analysis of RBP binding sites in RNA sequences.
• Post-transcriptional regulation studies: Investigation of mechanisms that control RNA stability, splicing, localization, and translation via RBP interactions.
• Gene expression regulation research: Linking RBP binding to changes in gene expression and regulatory networks.
• Disease mechanism exploration and therapeutic development: Studying RBP-related disease mechanisms and informing strategies that target specific RBPs.

Methodology:

Performs inexact k-mer spectrum search using Burrows-Wheeler Transform and FM-index to find significant seeds; evaluates seeds with a Deep Feed-forward Neural Network using flanking-region information and contextual features (pentamers, dinucleotide patterns, CG distribution); models are supported by molecular dynamics (MD) simulation studies and benchmarked on CLIP-seq-derived datasets (over 50 RBPs across three datasets, coverage of 131 RBPs).