Dengue Fever Prognosis Using Machine Learning

Authors

Kshitij Kadam – Texas A&M University (Department of Computer Science and Engineering)
Email: [email protected]
Adekola Okunola – Texas A&M University (Department of Electrical and Computer Engineering)
Email: [email protected]

Overview

This project aims to differentiate between Dengue Fever (DF) and Dengue Hemorrhagic Fever (DHF) using gene expression data. Advanced statistical and machine learning techniques, including ANOVA, Support Vector Machine (SVM), Random Forest Classifier, and Linear Discriminant Analysis (LDA), are used to classify patients accurately based on gene expression data.

Dataset

The dataset consists of gene expression profiles obtained from peripheral blood mononuclear cells (PBMCs) of 26 patients, covering 1981 genes. The classification is based on the following labels:

• DF (Dengue Fever)
• DHF (Dengue Hemorrhagic Fever)

The dataset reference is included in the final report.

Methodology

1. Data Preprocessing

• Load the dataset (Dengue_Fever_Prognosis_Dataset.csv).
• Identify columns corresponding to DHF and DF cases.
• Transpose the dataset for easier processing.
• Split the dataset into **training and testing sets.

2. Feature Selection with ANOVA

• Apply Analysis of Variance (ANOVA) to identify genes with significant differential expression.
• Select the top 10 genes with the lowest p-values.

3. Classification Models

a) Random Forest Classifier (See CP_Code1AI.py)

• Train a Random Forest model using the top selected genes.
• Evaluate the model's performance using accuracy score and classification report.
• Identify and visualize important features.

b) Support Vector Machine (SVM) (See CP_Code1AII.py & CP_Code1AV.py)

• Train SVM classifiers for different gene pairs.
• Visualize decision boundaries using scatter plots.
• Evaluate classifiers based on accuracy and misclassification analysis.

c) Linear Discriminant Analysis (LDA) (See CP_Code1AIII.py & CP_Code1AIV.py)

• Apply LDA to classify DF and DHF cases based on gene expression data.
• Evaluate performance metrics such as accuracy and misclassification analysis.
• Visualize LDA decision boundaries for selected gene pairs.

4. Model Evaluation

• The top classifiers are selected based on accuracy.
• Misclassified samples are analyzed to identify model limitations.
• Both SVM and LDA classifiers achieved high accuracy (1.0) on selected gene pairs.

Results

The analysis identified the most discriminatory genes and their impact on classifying DF vs. DHF. The models showed high accuracy and strong separation of classes, as demonstrated by the decision boundary visualizations.

Key Findings:

• ANOVA was successful in identifying genes with significant expression differences.
• SVM & LDA classifiers performed exceptionally well with 1.0 accuracy in top gene pairs.
• Random Forest identified critical gene importance for classification.

Project Files

File Name	Description
CP_Code1A.py	ANOVA-based feature selection for top discriminatory genes
CP_Code1AI.py	Random Forest Classifier for classification
CP_Code1AII.py	SVM model for individual gene classification
CP_Code1AIII.py	LDA classifier for gene pair analysis
CP_Code1AIV.py	LDA classifier with decision boundary visualization
CP_Code1AV.py	Advanced SVM classifier with iterative gene selection
Final Report.pdf	Detailed report covering research, methodology, results, and references

Programming Language, Tools, and Libraries

• Python
• Scikit-learn
• Numpy
• Matplotlib
• Pandas

Future Work

• Validate models with larger and more diverse datasets.
• Explore deep learning techniques for improved classification.
• Integrate findings into a real-time clinical decision-support system.

Acknowledgments

This was completed as a project at Texas A&M University. We appreciate the contributions of researchers in the field of bioinformatics and dengue prognosis.

References

@article{Nascimento2009,
    author = {Nascimento, E. and Abath, F. and Calzavara, C. and Gomes, A. and Acioli, B. and Brito, C. and Cordeiro, M. and Silva, A. and Andrade, C. M. R. and Gil, L. and Junior, U. B.-N. E. M.},
    title = {Gene expression profiling during early acute febrile stage of dengue infection can predict the disease outcome},
    journal = {PLoS ONE},
    volume = {4},
    number = {11},
    pages = {e7892},
    year = {2009},
    doi = {10.1371/journal.pone.0007892}
}

@book{BragaNeto2020,
    author = {Ulisses Braga-Neto},
    title = {Fundamentals of Pattern Recognition and Machine Learning},
    publisher = {Springer Nature Switzerland AG},
    year = {2020},
    isbn = {978-3-030-27655-3},
    doi = {10.1007/978-3-030-27656-0}
}

@misc{WHO2023,
    author = {{World Health Organization}},
    title = {Dengue and Severe Dengue: Fact Sheet},
    year = {2023},
    note = {Retrieved December 2, 2024, from \url{https://www.who.int/news-room/fact-sheets/detail/dengue-and-severe-dengue}}
}

@article{Liu2022,
    author = {Liu, Y. E. and Saul, S. and Rao, A. M. and others},
    title = {An 8-gene machine learning model improves clinical prediction of severe dengue progression},
    journal = {Genome Medicine},
    volume = {14},
    number = {33},
    year = {2022},
    doi = {10.1186/s13073-022-01034-w}
}