Random forests algorithm boosts genetic risk prediction of systemic lupus erythematosus

Area of Science:

• Genetics
• Computational Biology
• Immunology

Background:

• Systemic lupus erythematosus (SLE) presents diverse clinical symptoms, complicating diagnosis.
• Genetic factors play a significant role in SLE pathogenesis.
• Current polygenic risk scoring (PRS) models for SLE assume independent genetic variant contributions.

Purpose of the Study:

• To enhance the accuracy of SLE prediction by applying machine learning (ML) algorithms.
• To compare the performance of ML models against PRS for SLE classification.
• To identify the most effective ML approach for SLE genetic risk assessment.

Main Methods:

• Utilized data from genome-wide association studies (GWAS) in Chinese and European populations (19,208 participants).
• Applied and compared Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) algorithms.
• Assessed predictor performance using the Area Under the Receiver-Operator Curve (AUC).

Main Results:

• The RF model demonstrated superior performance in both Chinese (AUC=0.84) and European GWAS datasets.
• RF achieved a 13% improvement over the PRS model (AUC=0.74) in the Chinese cohort.
• RF achieved 84% sensitivity and 68% specificity for SLE classification at an optimal threshold.

Conclusions:

• Machine learning, particularly the RF model, offers a significant improvement over PRS for SLE prediction.
• The RF model shows consistent performance across different ethnic populations.
• The RF model represents a promising tool for the early diagnosis of SLE.