Combining Polygenic Risk Score and Voice Features to Detect Major Depressive Disorders

Abstract

Background: The application of polygenic risk scores (PRSs) in major depressive disorder (MDD) detection is constrained by its simplicity and uncertainty. One promising way to further extend its usability is fusion with other biomarkers. This study constructed an MDD biomarker by combining the PRS and voice features and evaluated their ability based on large clinical samples. Methods: We collected genome-wide sequences and utterances edited from clinical interview speech records from 3,580 women with recurrent MDD and 4,016 healthy people. Then, we constructed PRS as a gene biomarker by p value-based clumping and thresholding and extracted voice features using the i-vector method. Using logistic regression, we compared the ability of gene or voice biomarkers with the ability of both in combination for MDD detection. We also tested more machine learning models to further improve the detection capability. Results: With a p-value threshold of 0.005, the combined biomarker improved the area under the receiver operating characteristic curve (AUC) by 9.09% compared to that of genes only and 6.73% compared to that of voice only. Multilayer perceptron can further heighten the AUC by 3.6% compared to logistic regression, while support vector machine and random forests showed no better performance. Conclusion: The addition of voice biomarkers to genes can effectively improve the ability to detect MDD. The combination of PRS and voice biomarkers in MDD detection is feasible. This study provides a foundation for exploring the clinical application of genetic and voice biomarkers in the diagnosis of MDD.

Keywords: biomarkers; computer technology; depression; major depressive disorder (MDD); polygenic risk score (PRS); voice biomarkers.

FIGURE 1

Fivefold cross-validation of voice–gene data. In each fold, the samples were split into a training group and a test group. Voice and genetic sequence data of the training group were used to train the universal background model (UBM) and linear mixed model (LMM) separately. Then, i-vectors for the training and test groups were extracted through the UBM, and the polygenic risk score (PRS) can be calculated through the LMM. The i-vectors and PRS will be concatenated as input features for a machine learning (ML) model.

FIGURE 2

Process of i-vector extraction. UBM-GMM is a universal background model adapted by a Gaussian mixture model. n = 256 means there were 256 Gaussian mixture clusters. d = 400 means the dimension of i-vectors is 400.

FIGURE 3

Polygenic risk score (PRS) model prediction results with different p-value thresholds (PTs) under different covariate use strategies. no-cov, no covariates were considered during the training and prediction processes; all-cov, all covariates were considered during the training and prediction processes; random-cov, the PRS model was trained with a sample genetic matrix along with covariates, but made predictions on samples whose covariates were replaced with random numbers. AUC, Area under the receiver operating characteristic curve.

FIGURE 4

Prediction results with different p-value thresholds (PTs) using different biomarkers. The x-axis is the p-value threshold (PT) used in the gene model and the combined biomarkers. Voice biomarkers are not related to PT and are indicated by a dashed horizontal line. AUC, Area under the receiver operating characteristic curve.

FIGURE 5

Stratified population accuracy using different biomarkers. The test samples were divided into three groups according to their predicted polygenic risk scores (PRSs). Accuracies were calculated for the three groups separately.