Using amino acids co-occurrence matrices and explainability model to investigate patterns in dengue virus proteins

Area of Science:

• Virology
• Genomics
• Bioinformatics

Background:

• Dengue virus (DENV) causes vector-borne diseases with mild to severe symptoms.
• Severe dengue outcomes are multifactorial and not fully understood.
• Viral genetic signatures may influence disease severity.

Purpose of the Study:

• To identify genetic markers associated with severe dengue.
• To analyze patterns in dengue virus protein sequences for severe illness prediction.

Main Methods:

• Mapping protein sequences to matrices using amino acid co-occurrence.
• Classifying proteins associated with severe dengue using Random Forests.
• Interpreting models with SHAP Values to find key amino acid patterns.

Main Results:

• Ten binary classifiers were trained for each DENV protein.
• DENV protein E showed the highest performance across five metrics (PR-AUC, ROC-AUC, F1-score, Precision, Recall).
• Amino acid co-occurrence patterns in Domain 1 of protein E correlated with severe dengue outcomes.

Conclusions:

• Specific amino acid co-occurrence patterns in DENV protein E are strongly associated with severe dengue.
• These patterns, particularly in Domain 1, can aid in predicting severe illness.
• Findings may inform new dengue treatments and theoretical discussions.