DeepCLIP

DeepCLIP predicts the impact of nucleotide variants on protein–RNA binding interactions using a deep learning model that operates on primary sequence data to enable context-aware binding predictions.

Key Features:

• Sequence-only input: Operates exclusively on primary nucleotide sequence data for modeling protein–RNA interactions.
• Deep learning architecture: Employs a neural network combining shallow convolutional layers with a bidirectional Long Short-Term Memory (LSTM) layer.
• Context-aware modeling: Captures both local sequence motifs and broader genomic context that influence RNA-binding protein specificity.
• Variant effect prediction: Predicts the impact of nucleotide variants on protein–RNA binding interactions.
• Binding profiles: Generates binding profiles that can be used to evaluate positional effects on binding and guide antisense oligonucleotide design.
• Benchmarking and experimental correlation: Demonstrated to surpass existing methods and to show strong correlation with independent wet lab functional outcomes.

Scientific Applications:

• Antisense oligonucleotide design: Uses binding profiles to inform design of antisense oligonucleotides for therapeutic targeting.
• Variant interpretation: Assesses how nucleotide variants affect protein–RNA binding relevant to disease mechanisms.
• Regulatory mechanism analysis: Investigates position-dependent regulation of RNA–protein interactions in a tissue-specific manner.
• RNA processing studies: Studies effects on splicing, nuclear shuttling, and transcript stability through predicted binding changes.

Methodology:

DeepCLIP uses a neural network combining shallow convolutional layers and a bidirectional Long Short-Term Memory (LSTM) layer trained on primary nucleotide sequence data to perform context-aware predictions of protein–RNA binding.