Pharmacogenomics‐Driven Prediction of Antidepressant Treatment Outcomes: A Machine‐Learning Approach With Multi‐trial Replication

Abstract

We set out to determine whether machine learning–based algorithms that included functionally validated pharmacogenomic biomarkers joined with clinical measures could predict selective serotonin reuptake inhibitor (SSRI) remission/response in patients with major depressive disorder (MDD). We studied 1,030 white outpatients with MDD treated with citalopram/escitalopram in the Mayo Clinic Pharmacogenomics Research Network Antidepressant Medication Pharmacogenomic Study (PGRN‐AMPS; n = 398), Sequenced Treatment Alternatives to Relieve Depression (STAR*D; n = 467), and International SSRI Pharmacogenomics Consortium (ISPC; n = 165) trials. A genomewide association study for PGRN‐AMPS plasma metabolites associated with SSRI response (serotonin) and baseline MDD severity (kynurenine) identified single nucleotide polymorphisms (SNPs) in DEFB1, ERICH3, AHR, and TSPAN5 that we tested as predictors. Supervised machine‐learning methods trained using SNPs and total baseline depression scores predicted remission and response at 8 weeks with area under the receiver operating curve (AUC) > 0.7 (P < 0.04) in PGRN‐AMPS patients, with comparable prediction accuracies > 69% (P ≤ 0.07) in STAR*D and ISPC. These results demonstrate that machine learning can achieve accurate and, importantly, replicable prediction of SSRI therapy response using total baseline depression severity combined with pharmacogenomic biomarkers.

Precision Psychiatry Applications with Pharmacogenomics: Artificial Intelligence and Machine Learning Approaches

“…Athreya et al [30] employed a random forest method, a conventional artificial intelligence and machine learning approach, to predict antidepressant therapy response (with AUC > 0.7 and accuracy > 69%). The multi-omics data in their study included the SNPs datasets.…”

“…To predict antidepressant treatment response, various research studies [22][23][24][25][26][27][28][29][30] used conventional artificial intelligence and machine learning methods. For example, Kautzky et al [22] suggested that the random forest structure, a conventional artificial intelligence and machine learning method, correctly predicted 25% of responders for antidepressant treatment outcome based on genetic and clinical data.…”

Multiscale Modeling Meets Machine Learning: What Can We Learn?

“…Classification is a supervised learning approach in which the algorithm learns from a training set of correctly classified observations and uses this learning to classify new observations, where the output variable is discrete. Examples in biomedicine include classifying whether a tumor is benign or malignant [34], classifying the effects of individual single nucleotide polymorphisms on depression [7], the effects of ion channel blockage on arrhythmogenic risk in drug development [108], and the effects of chemotherapeutic agents in personalized cancer medicine [27].…”

“…In biomedicine, the original area of big-data research was the identification of the human genome, which employed machine learning to construct consistent frames out of genome fragments. Since then, extensions of genomic studies involve large numbers of patients and controls in genome-wide association studies, which compare single nucleotide polymorphisms in patients versus controls [7,56]. Single nucleotide polymorphisms adjacent to coding sequences suggest which gene product might be involved in the disease.…”

Who Believes They Are “Just Average”: Informing the Treatment of Individual Patients Using Population Data

“…The challenge is to collect data reflecting the uniqueness of an individual's genome, metabolome, microbiome, current state of health (including pre‐existing medical conditions and concurrent medications), lifestyle (diet, stress level, physical activity, and sleep) on many people—a population—and allow machine‐learning algorithms to select the features most relevant and informative for a particular situation. Examples of artificial intelligence successfully applied to predict drug response are encouraging, but successful clinical implementation on a population basis will require further evidence from prospective clinical trials that the method performs as intended in a broader range of patients than originally studied.…”