A Novel Feature Extraction Method with Feature Selection to Identify Golgi-Resident Protein Types from Imbalanced...

Area of Science:

• Molecular Biology
• Bioinformatics
• Cell Biology

Background:

• The Golgi Apparatus (GA) is crucial for protein processing and transport; its protein dysfunction is linked to neurodegenerative diseases.
• Accurate identification of protein localization within the GA is vital for drug development and understanding cellular mechanisms.
• Distinguishing between cis-Golgi and trans-Golgi proteins is essential for elucidating GA function.

Purpose of the Study:

• To develop a novel computational method for differentiating cis-Golgi from trans-Golgi proteins using protein sequence data.
• To enhance the accuracy of protein sub-Golgi localization prediction.
• To provide a tool for investigating Golgi-related cellular processes and diseases.

Main Methods:

• A new feature extraction technique based on Common Spatial Patterns (CSP) was developed to capture evolutionary information from protein sequences.
• The Synthetic Minority Over-sampling Technique (SMOTE) was used to address dataset imbalance.
• A Random Forest (RF) classifier, guided by Random Forest-Recursive Feature Elimination (RF-RFE) for feature selection, was employed.

Main Results:

• The proposed method achieved high performance in distinguishing cis-Golgi from trans-Golgi proteins, with an accuracy of 0.885 and MCC of 0.765 via jackknife cross-validation.
• The method demonstrated significantly improved performance on an independent dataset.
• CSP-based feature extraction showed promise for predicting protein function and localization.

Conclusions:

• The developed computational method offers a robust and accurate approach for identifying Golgi-resident protein types.
• This technique can aid in understanding the molecular mechanisms underlying neurodegenerative diseases linked to Golgi dysfunction.
• The CSP feature extraction method holds potential for broader applications in protein function prediction.