Building a genetic risk model for bipolar disorder from genome-wide association data with random forest algorithm

Area of Science:

• Genetics
• Psychiatry
• Machine Learning

Background:

• Complex diseases with high heritability, such as bipolar disorder (BPD), can benefit from genetic risk scores for clinical diagnosis.
• Large-scale genome-wide association (GWA) data provides a foundation for developing predictive genetic models.

Purpose of the Study:

• To construct a genetic risk model for bipolar disorder (BPD) using a machine learning approach.
• To identify informative genetic markers capable of differentiating BPD from healthy individuals.

Main Methods:

• Utilized GWA data from the Genetic Association Information Network as training data and Systematic Treatment Enhancement Program (STEP) GWA data for validation.
• Applied a random forest algorithm to pre-filtered markers and assessed variable importance.
• Selected 289 candidate markers based on random forest procedures.

Main Results:

• The random forest model achieved an area under the receiver operating characteristic curve (AUC) of 0.944 in the training set and 0.702 in the validation set.
• A genetic risk score with a cutoff of 184 demonstrated a sensitivity of 0.777 and specificity of 0.854 for BPD.
• Pathway analyses identified significant biological pathways associated with the selected genes.

Conclusions:

• The study successfully identified genetic markers with acceptable discriminability for differentiating BPD from healthy controls in the validation dataset.
• Future research can enhance diagnostic classification by integrating comprehensive clinical risk factors with genetic data in larger sample sizes.