PlncRNA-HDeep

PlncRNA-HDeep predicts plant long noncoding RNAs (lncRNAs) using a hybrid deep learning approach that integrates p-nucleotide and one-hot encodings to characterize RNA sequences.

Key Features:

• Hybrid Deep Learning Model: Integrates p-nucleotide and one-hot encodings to capture diverse information present in RNA sequences.
• Ensemble Approach: Employs Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) trained separately—treating sequences as sentences and images—and hybridizes them into an ensemble.
• Sequence-only Input: Operates solely on RNA sequence data without requiring explicit feature engineering or prior biological knowledge.
• High Predictive Performance: Outperforms single-model approaches (lncRNA-LSTM and CNN), shallow machine learning methods (support vector machines, random forests, k-nearest neighbors, decision trees, naive Bayes, logistic regression), and tools (CNCI, PLEK, CPC2, LncADeep, lncRNAnet), achieving 97.9% sensitivity, 95.1% precision, 96.5% accuracy, and 96.5% F1 on the Zea mays dataset.
• Optimized Parameters and Hybrid Strategies: Model parameters were adjusted and three hybrid strategies were tested to maximize predictive performance.

Scientific Applications:

• Plant genomics and molecular biology: Facilitates discovery and characterization of plant lncRNAs and exploration of their roles in regulating biological activities.

Methodology:

Sequences are encoded using p-nucleotide and one-hot schemes, LSTM and CNN models are trained separately (treating sequences as sentences and images) and then hybridized into an ensemble, with parameters optimized and three hybrid strategies evaluated.