A Random Forest Sub-Golgi Protein Classifier Optimized via Dipeptide and Amino Acid Composition Features

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics and Computational Biology
• Cell Biology

Background:

• The Golgi apparatus plays a crucial role in cellular function, and its malfunction is linked to genetic and neurodegenerative diseases.
• Accurate identification of sub-Golgi proteins (cis-Golgi and trans-Golgi) is essential for understanding these diseases.

Purpose of the Study:

• To develop a state-of-the-art computational tool for classifying sub-Golgi proteins.
• To provide a practical and efficient method for identifying Golgi protein localization without requiring sequence alignment.

Main Methods:

• Development of a random forests classifier named rfGPT.
• Utilized 2-gap dipeptide and split amino acid composition for feature vectors.
• Employed synthetic minority over-sampling technique (SMOTE) and analysis of variance (ANOVA) for feature selection.

Main Results:

• The optimal rfGPT classifier achieved high performance metrics, including 90.5% accuracy (ACC), 0.811 Matthews correlation coefficient (MCC), 92.6% sensitivity (Sn), and 88.4% specificity (Sp) during training.
• Independent testing yielded ACC = 90.6%, MCC = 0.696, Sn = 96.1%, and Sp = 69.2%.
• rfGPT is the leading sub-Golgi protein predictor using feature vectors without position-specific scoring matrices, making it practical for large-scale protein sequence analysis.

Conclusions:

• rfGPT demonstrates superior independent testing scores among Golgi classifiers optimized on smaller datasets.
• The classifier's reliance on sequence composition features makes it highly practical, eliminating the need for sequence alignment.
• Key biological features for distinguishing sub-Golgi proteins include non-polar/aliphatic residues composition and specific dipeptide compositions.