Protein Sub-Nuclear Localization Based on Effective Fusion Representations and Dimension Reduction Algorithm LDA.

Area of Science:

• Bioinformatics
• Computational Biology
• Proteomics

Background:

• Effective protein sequence representation is vital for predicting sub-nuclear localization.
• Existing methods like dipeptide composition (DipC), pseudo-amino acid composition (PseAAC), and position-specific scoring matrices (PSSM) have limitations due to their single-perspective nature.

Purpose of the Study:

• To develop novel, integrated feature representations for protein sequences.
• To enhance the accuracy of protein sub-nuclear localization prediction by combining complementary sequence information.

Main Methods:

• Proposed two fusion feature representations: DipPSSM (integrating PSSM with DipC) and PseAAPSSM (integrating PSSM with PseAAC).
• Introduced balance factors to weigh component importance, optimized using a genetic algorithm.
• Applied Linear Discriminant Analysis (LDA) for dimensionality reduction of high-dimensional fused features.
• Evaluated performance using KNN classifier and cross-validation on two public datasets.

Main Results:

• The proposed fused representations (DipPSSM and PseAAPSSM) demonstrated superior performance over traditional methods for protein sub-nuclear localization.
• LDA-treated representations showed improved prediction accuracy compared to untreated ones.
• The study achieved higher sensitivity, specificity, accuracy, and MCC scores with the novel methods.

Conclusions:

• Fusion of PSSM with DipC and PseAAC offers a more comprehensive protein sequence representation.
• Optimized fusion strategies and dimensionality reduction techniques like LDA are crucial for improving sub-nuclear localization prediction.
• The developed methods provide a significant advancement in predicting protein subcellular locations.