KD-KLNMF

KD-KLNMF predicts the subcellular localization of long non-coding RNAs (lncRNAs) using k-mer and dinucleotide-based spatial autocorrelation features combined with Kullback-Leibler divergence-based nonnegative matrix factorization and Support Vector Machine classification.

Key Features:

• Feature Vector Construction: Uses k-mer and dinucleotide-based spatial autocorrelation to capture sequence composition and positional dependencies of lncRNAs.
• Handling Imbalanced Datasets: Applies the Synthetic Minority Over-sampling Technique (SMOTE) to balance class distributions during preprocessing.
• Feature Selection via Nonnegative Matrix Factorization: Employs Kullback-Leibler divergence-based nonnegative matrix factorization for dimensionality reduction and informative feature selection.
• Classification with Support Vector Machine: Uses a Support Vector Machine (SVM) classifier selected after comparison with other classifiers to assign subcellular localization labels.
• Model Evaluation: Validates performance using the jackknife test and reports accuracies of 97.24% on the training set and 92.86% on an independent dataset.

Scientific Applications:

• lncRNA localization inference: Predicts subcellular compartments of lncRNAs to inform hypotheses about their cellular roles.
• Functional annotation of lncRNAs: Supports inference of lncRNA functions based on predicted localization patterns.
• Genomics and transcriptomics research: Provides localization data useful for integrative analyses in genomics and transcriptomics studies.
• Systems biology and disease studies: Facilitates incorporation of lncRNA localization into systems-level models and investigations of disease mechanisms.

Methodology:

Feature extraction via k-mer and dinucleotide-based spatial autocorrelation; data preprocessing with SMOTE; dimensionality reduction and feature selection using Kullback-Leibler divergence-based nonnegative matrix factorization; classification with SVM; and evaluation by jackknife test.