Machine Learning Data Imputation and Classification in a Multicohort Hypertension Clinical Study.

Area of Science:

• Genomics
• Translational Research
• Medical Informatics

Background:

• Healthcare initiatives promote clinical data use for medical discovery.
• Machine learning (ML) aids in detecting patterns in complex diseases like hypertension.
• Previous genomic studies in African Americans (AA) focused on rare variants for hypertension, yielding limited results.

Purpose of the Study:

• To apply ML for analyzing phenotype data in African American (AA) participants within the Minority Health Genomics and Translational Research Repository Database.
• To impute missing phenotype data using neural networks to expand the usable clinical dataset.
• To validate the expanded dataset's utility for identifying associations between phenotype variables and hypertension case/control status.

Main Methods:

• Utilized neural networks for phenotype data imputation to address missing values.
• Expanded the clinical dataset size through data imputation.
• Employed data mining classification tools to generate association rules.

Main Results:

• The expanded dataset, created by ML imputation, demonstrated improved performance in associating phenotype variables with hypertension status.
• Association rules were successfully generated using data mining techniques.
• The study highlights the effectiveness of ML in uncovering complex relationships in genomic and clinical data.

Conclusions:

• Machine learning imputation is effective for increasing the usability of clinical datasets for hypertension research in African American populations.
• This approach can uncover novel associations between phenotype and genotype data, advancing translational research.
• The findings support the use of advanced statistical methods for complex disease research in diverse populations.