Using LASSO regression to detect predictive aggregate effects in genetic studies.

Area of Science:

• Genetics
• Biostatistics
• Machine Learning

Background:

• Risk prediction models are crucial for understanding disease etiology.
• Identifying informative genetic markers and phenotypic features is essential for accurate prediction.

Purpose of the Study:

• To evaluate different strategies for applying LASSO regression in genetic risk prediction.
• To compare the performance of LASSO models using only genetic data versus combined genetic and phenotypic data.

Main Methods:

• Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed.
• A 5-fold cross-validation approach was used on the Genetic Analysis Workshop 17 exome simulation data.
• Model performance was assessed using the area under the receiver operating curve (AUC).

Main Results:

• LASSO models using only genotypic markers achieved AUCs ranging from 0.45 to 0.63.
• Incorporating both genotypic and phenotypic features (smoking, age, sex) improved AUCs to 0.69-0.87.
• LASSO demonstrated limited ability to identify true causal markers but could detect some common variants like FLT1.

Conclusions:

• Combining genetic and phenotypic data substantially enhances the predictive power of LASSO models.
• LASSO is a valuable tool for feature selection in genetic risk prediction, although its causal marker identification capabilities are constrained.
• Further research may refine LASSO strategies for improved genetic marker discovery.