Biomarker discovery for predicting spontaneous preterm birth from gene expression data by regularized logistic...

Area of Science:

• Computational biology
• Genomics
• Biostatistics

Background:

• Spontaneous preterm birth (SPTB) poses significant risks to infants and mothers.
• Accurate identification of SPTB biomarkers is crucial for early intervention and risk reduction.
• Existing feature selection methods for gene expression data have limitations.

Purpose of the Study:

• To develop and evaluate a computational method for discovering SPTB biomarkers using regularized logistic regression.
• To compare the performance of seven different regularization penalties for gene selection.
• To construct a predictive model for SPTB risk assessment.

Main Methods:

• Applied regularized logistic regression with seven distinct penalties (including elastic net, lasso, and SCAD) to gene expression data.
• Performed functional enrichment analysis on identified biomarkers.
• Developed a logistic regression classifier to generate a preterm risk score (PRS).
• Validated the identified biomarkers on an independent dataset.

Main Results:

• Elastic net, lasso, and SCAD penalties demonstrated superior performance in identifying SPTB biomarkers.
• The developed classifier achieved an Area Under the Curve (AUC) of 0.933 for SPTB classification on an independent dataset.
• Functional enrichment analysis provided biological context for the identified biomarkers.

Conclusions:

• Regularized logistic regression is an effective strategy for discovering SPTB biomarkers from gene expression data.
• The proposed method and identified biomarkers show high accuracy in predicting SPTB risk.
• This approach is adaptable for biomarker discovery in other complex diseases.