A growing body of evidence now suggests that precision psychiatry, an interdisciplinary field of psychiatry, precision medicine, and pharmacogenomics, serves as an indispensable foundation of medical practices by offering the accurate medication with the accurate dose at the accurate time to patients with psychiatric disorders. In light of the latest advancements in artificial intelligence and machine learning techniques, numerous biomarkers and genetic loci associated with psychiatric diseases and relevant treatments are being discovered in precision psychiatry research by employing neuroimaging and multi-omics. In this review, we focus on the latest developments for precision psychiatry research using artificial intelligence and machine learning approaches, such as deep learning and neural network algorithms, together with multi-omics and neuroimaging data. Firstly, we review precision psychiatry and pharmacogenomics studies that leverage various artificial intelligence and machine learning techniques to assess treatment prediction, prognosis prediction, diagnosis prediction, and the detection of potential biomarkers. In addition, we describe potential biomarkers and genetic loci that have been discovered to be associated with psychiatric diseases and relevant treatments. Moreover, we outline the limitations in regard to the previous precision psychiatry and pharmacogenomics studies. Finally, we present a discussion of directions and challenges for future research.
Keywords: artificial intelligence, biomarker, deep learning, machine learning, multi-omics, neural networks, neuroimaging, pharmacogenomics, precision medicine, precision psychiatry
Precision psychiatry, which is an emerging interdisciplinary field of psychiatry, precision medicine, and pharmacogenomics, is developing into essential practices in medicine with a promise of the individualization of clinical care for patients with psychiatric disorders. It should be pointed out that pharmacogenomics is sometimes used interchangeably with precision medicine. In other words, pharmacogenomics is one of the research fields to further advance precision psychiatry, where pharmacogenomics is defined as the study of how genes and their functions can influence a person’s response to medications. In general, precision psychiatry means that medical decisions, treatments, and practices are adapted to specific patients with psychiatric disorders . More specifically, the whole population of patients with psychiatric disorders are separated into various groups by specific biomarkers, where multi-omics and/or neuroimaging datasets are available to represent these specific biomarkers within the context of precision psychiatry. Therefore, medications can be tailored individually to each respective patient by using pertinent or proportionate genetic biomarkers and/or imaging attributes . To date, there are progressively growing genetic biomarkers and imaging attributes that may contribute to the prognosis and treatment response for patients with psychiatric disorders . For instance, it has long been recognized that genetic biomarkers such as gene expression profiles and single nucleotide polymorphisms (SNPs) can be utilized to evaluate adverse drug reactions and clinical treatment response for antidepressants in patients with major depressive disorder (MDD) [3,4].
In the field of precision psychiatry, researchers integrate multiple data types such as multi-omics and neuroimaging data with state-of-the-art artificial intelligence and machine learning algorithms, which can accordingly learn to identify complex patterns with respect to observational datasets [5,6,7]. Namely, multi-omics and neuroimaging data are employed to serve as biomarkers (or predictive factors) to fulfill the concept of precision psychiatry by using artificial intelligence and machine learning algorithms. In order to address the demanding challenges we face today in the field of precision psychiatry, there is an enormous need for developing software tools in artificial intelligence and machine learning frameworks that can predict specific quantitative and/or categorical phenotypes in clinical settings by utilizing next-generation multi-omics and neuroimaging datasets [5,6,7].
In recent times, scientists have been making significant progress in the multidisciplinary fields of artificial intelligence, machine learning, precision psychiatry, pharmacogenomics, multi-omics, and neuroimaging [5,6,7]. The goal of an artificial intelligence and machine learning approach is to provide a data-driven algorithm that can in general learn from the data in the past and/or in the present by leveraging the learned insight into estimating predictive outcomes for any unknown data and/or for any unknown event in the future [8,9,10]. In the general terms, the guideline of an artificial intelligence and machine learning approach is comprised of the following three steps: we firstly build the predictive model from the initial input data in the beginning step, then secondly fine-tune and gauge the predictive model in the intermediate step, and thirdly utilize the predictive model for presenting an estimated outcome in the final step [8,9,10].
Latest advancements in artificial intelligence and machine learning technologies, especially deep learning algorithms, have revealed their promising capacities to recognize and learn complex and nonlinear hierarchical patterns with respect to mammoth large-scale experimental data [11,12,13,14,15]. Furthermore, deep learning algorithms have accomplished state-of-the-art performances on a wide range of medical applications such as precision psychiatry based on recent new technologies such as the invention of general-purpose computing on graphics processing units [6,12,13,14,15]. Primarily, the objective of deep learning algorithms is to build artificial intelligence and machine learning algorithms which employ multiple layers of abstraction such as artificial neural networks to construct a hierarchical representation for the data [12,13,14,15,16]. That is, deep learning algorithms for classification applications, such as diagnosis prediction in precision psychiatry, are procedures for determining the best hypothesis by utilizing artificial neural networks with multiple layers, instead of utilizing artificial neural networks with only one single layer [12,13,14,15,16].
With the recent advance in multi-omics and neuroimaging technologies, precision psychiatry shows high growth potential to respond to the needs of new diagnostic tools as well as novel drugs for treatment and therapeutic interventions . Furthermore, the usage of biomarkers has played a key role in precision psychiatry based on artificial intelligence and machine learning approaches . In the recent past, there were a wide variety of emerging vital research studies for numerous diseases and treatments of significance for precision psychiatry with consideration of artificial intelligence and machine learning methods . Accordingly, it could be remarkably intriguing to design artificial intelligence and machine learning algorithms that can predict the potential outcomes of drug treatments and disease status for patients with psychiatric disorders [6,17]. To address this challenge, artificial intelligence and machine learning approaches may yield helpful software tools to achieve the promise of precision psychiatry by concerning specific biomarkers for drug treatments and disease status [6,17]. In this review, we show recent research studies in precision psychiatry and pharmacogenomics, which assessed disease status and drug treatments using artificial intelligence and machine learning approaches, such as deep learning and artificial neural network algorithms. We mainly focus on multi-omics and neuroimaging data. Additionally, we present the limitations in these research studies and summarize a discussion of future challenges as well as directions.
Here, in the context of artificial intelligence and machine learning methods, we provide various research studies with focus on four major categories including treatment prediction, prognosis prediction, diagnosis prediction, and the detection of potential biomarkers in terms of psychiatric disorders, precision psychiatry, and pharmacogenomics. Biological and/or clinical implications from these four major arenas can serve as decision support aides for treatment prediction, prognosis prediction, and diagnosis prediction in translational and precision psychiatry . Furthermore, we particularly focus on psychiatric disorders such as MDD, bipolar disorder, attention deficit hyperactivity disorder, Alzheimer’s disease, autism spectrum disorder, and schizophrenia. While this review does not support the full set of related research studies reported in the literature, it nonetheless describes a synthesis of those that can markedly influence public and population health-oriented applications in psychiatric disorders, precision psychiatry, and pharmacogenomics in the near to mid-term future.
The usage of artificial intelligence and machine learning methods is still in its infancy in terms of forecasting drug treatments in psychiatric drugs due to the fact that scant human studies have explored the predictive models of evaluating drug treatment response. Here, we focus on antidepressant and lithium treatment outcomes using deep learning strategies as well as conventional artificial intelligence and machine learning methods in this section (Table 1). In this review, we first conducted a comprehensive search of the electronic PubMed database (2015-present) using key words such as “machine learning,” “deep learning,” “antidepressant,” “lithium,” “major depressive disorder,” “bipolar disorder,” and “pharmacogenomics”. Then, we manually screened the obtained articles with a particular focus on MDD. The multi-omics data in these selected studies included SNPs datasets, DNA methylation datasets, gene expression datasets, and phenotypic datasets (such as demographic and clinical datasets). In addition, the reader can refer to a recent review by Pisanu and Squassina  for treatment-resistant schizophrenia, where patients with treatment-resistant schizophrenia are defined as those revealing little or no response to at least two non-clozapine antipsychotic medications. The reader can also refer to a recent review by Perlman et al.  for predictors of antidepressant treatment response in MDD, where predictor categories include demographic factors, symptom profiles (such as age of onset), peripheral markers (accessible through urine, blood, or saliva), genetic biomarkers, and neuroimaging data.
Relevant studies on the predictive models of evaluating drug treatment response.
|Lin et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B21-ijms-21-00969]||Deep learning architecture||AUC = 0.82, sensitivity = 0.75, specificity = 0.69 for antidepressant treatment response;AUC = 0.81, sensitivity = 0.77, specificity = 0.66 for remission|
|Kautzky et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B22-ijms-21-00969]||Random forest||An accuracy of 25% for antidepressant treatment outcome|
|Patel et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B23-ijms-21-00969]||Decision tree||An accuracy of 89% based on mini-mental status examination scores, age, and structural imaging|
|Chekroud et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B24-ijms-21-00969]||Tree-based ensemble||An accuracy of 59% based on 25 variables for clinical antidepressant remission|
|Iniesta et al. also [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B25-ijms-21-00969]||Elastic net||AUC = 0.72 based on clinical and demographical datasets|
|Maciukiewicz et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B27-ijms-21-00969]||SVM and decision trees||An accuracy of 52% based on SNPs|
|Chang et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B29-ijms-21-00969]||Linear regression||An accuracy of 84% based on neuroimaging biomarkers, genetic variants, DNA methylation, and demographic information|
|Athreya et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B30-ijms-21-00969]||Random forest||AUC > 0.7 and accuracy > 69% for antidepressant therapy response|
|Nunes et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B31-ijms-21-00969]||Random forest||AUC = 0.8; sensitivity = 0.53; specificity = 0.9 for lithium therapy response|
|Eugene et al. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037937/#B32-ijms-21-00969]||Decision tree and random forest||AUC = 0.92 for lithium therapy response|