MFF-HPO: Protein-Phenotype Associations Prediction Based on Sequence Using Multi-Feature Fusion

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Protein abnormalities are linked to cellular dysfunction and disease development.
• Identifying disease-associated proteins is vital for precision medicine.
• Current computational methods lack deep feature extraction from protein sequences and structures.

Purpose of the Study:

• To develop a novel computational model, MFF-HPO, for predicting protein-phenotype associations.
• To overcome limitations of existing methods by integrating multi-view features.
• To enhance the identification of disease-associated proteins.

Main Methods:

• Generating 3D protein structures from amino acid sequences.
• Deriving contact graphs and secondary structures.
• Integrating sequence encoding and physicochemical properties.
• Utilizing Graph Attention Networks and deep neural networks for feature extraction.
• Fusing multi-view features for phenotype prediction.

Main Results:

• MFF-HPO demonstrated superior performance compared to state-of-the-art methods.
• Achieved a mean area under the precision-recall curve of 0.314.
• Obtained a mean Fmax score of 0.371.
• Ablation studies confirmed the benefit of multi-view feature fusion.

Conclusions:

• MFF-HPO effectively predicts protein-phenotype associations by integrating diverse features.
• The model's multi-view feature fusion enhances predictive accuracy.
• MFF-HPO offers a practical and efficient approach for identifying disease-associated proteins.