HPOseq: a deep ensemble model for predicting the protein-phenotype relationships based on protein sequences

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Understanding protein-phenotype relationships is crucial for disease detection and personalized medicine.
• Large-scale proteomics data aids this understanding, but computational methods face challenges due to limited protein information.
• Existing computational methods show promise but require further refinement for efficiency and cost-effectiveness.

Purpose of the Study:

• To develop an accurate computational model for predicting human protein-phenotype relationships.
• To leverage only protein sequence information for prediction, reducing reliance on external data.
• To improve the efficiency and reduce the cost of identifying disease-related proteins.

Main Methods:

• An ensemble prediction model, HPOseq, was developed.
• Two base models were established: one using amino acid sequence features, the other using protein-protein network information derived from sequence similarity.
• An ensemble module integrated predictions from both base models.

Main Results:

• HPOseq accurately predicts human protein-phenotype relationships based solely on sequence information.
• The model integrates internal protein sequence features and inter-protein network information for enhanced prediction.
• Case studies confirmed the practical significance of HPOseq in phenotype annotation and protein analysis.

Conclusions:

• HPOseq outperforms seven baseline methods in predicting protein-phenotype relationships, as shown by 5-fold cross-validation.
• The model's reliance on sequence information makes it a valuable tool for biological discovery.
• HPOseq demonstrates practical utility and significance in biological research applications.