lncLocator-2.0

lncLocator-2.0 predicts the subcellular localization of long non-coding RNAs (lncRNAs) for individual cell lines to capture tissue-specific expression and localization differences.

Key Features:

• Cell-Line-Specific Prediction: Trains distinct deep learning models for each of 15 benchmarked cell lines to provide cell-line-specific localization predictions.
• End-to-End Deep Learning Model: Integrates convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and multilayer perceptrons (MLPs) in an end-to-end framework to classify lncRNA subcellular localizations.
• Word Embeddings from Natural Language Models: Employs word embeddings learned through natural language processing techniques to represent nucleotide sequence patterns.
• Interpretable Predictions with Integrated Gradients: Applies Integrated Gradients to attribute predictions to nucleotide patterns and provide model interpretability.

Scientific Applications:

• Research on Tissue-Specific Expression: Supports investigation of how lncRNA localization varies across cell lines and tissues.
• Understanding lncRNA Functionality: Aids linking subcellular localization to nucleotide sequence features to infer potential lncRNA regulatory roles.
• Development of Targeted Therapies: Informs strategies that leverage cell-line-specific lncRNA localization for therapeutic targeting.

Methodology:

Constructs benchmark datasets for 15 distinct cell lines, trains end-to-end deep learning models (CNNs, LSTMs, MLPs) on sequence data using learned word embeddings, and uses Integrated Gradients for prediction interpretability.