LPI-CNNCP

LPI-CNNCP predicts lncRNA–protein interactions using a convolutional neural network that applies a copy-padding trick and high-order one-hot encoding to convert variable-length RNA and protein sequences into fixed-length, image-like inputs for predictive modeling.

Key Features:

• Target biomolecules: Predicts interactions between long noncoding RNAs (lncRNAs) and RNA binding proteins (RBPs).
• Copy-padding trick: Converts variable-length protein and RNA sequences into fixed-length sequences to satisfy CNN input requirements.
• High-order one-hot encoding: Transforms amino acid and nucleotide sequences into image-like representations that capture dependencies among residues.
• Convolutional neural network: Processes encoded sequence images for interaction prediction.
• Evaluation protocol: Performance assessed using 10-fold cross-validation (10CV) and independent test sets.
• Comparative benchmarking: Copy-padding was compared against zero-padding and cropping and was reported to outperform both methods.
• Performance against state-of-the-art: Demonstrated superior performance relative to other methods in comparative tests described.

Scientific Applications:

• Interaction prediction: Identifies potential lncRNA–protein (lncRNA–RBP) interactions.
• Functional inference: Supports investigation of lncRNA roles in post-transcriptional regulation, cell differentiation, and gene regulation via predicted RBP partners.
• Method benchmarking: Provides a framework for comparing sequence encoding and padding strategies in machine learning approaches for biological sequences.

Methodology:

Variable-length sequences are converted to fixed-length using the copy-padding trick, sequences are encoded via high-order one-hot encoding into image-like inputs, those inputs are processed by a convolutional neural network, and performance is evaluated with 10-fold cross-validation and independent tests including comparisons to zero-padding and cropping.