Digital Sensory Phenotyping for Psychiatric Disorders

Today’s genome-wide association studies (GWAS) of psychiatric disorders require massive sample sizes and the identification of biologically relevant phenotypes. Sensory phenotypes, assessed by measuring sensorial function, represent early symptoms of psychiatric disorders, and may involve neurobiological pathways in psychiatric disorders. Yet, sensory phenotypes have rarely been studied in large populations for early diagnosis or GWAS. The concept of using digital devices to collect data on disease-related phenotypes is beginning to attract considerable attention. Is it possible to assess sensory phenotypes dynamically by digital devices? And furthermore, is it possible to explain the pathology of psychiatric disorders through those assessments? In this review, we summarize studies investigating sensory phenotypes and digital phenotyping of psychiatric disorders. We discuss the feasibility of digital phenotyping to better capture disease-related sensory phenotypes. We also discussed potential ethical and privacy issues, which require regulation of governments and collaborations of all researchers to solve. While the emergence of digital phenotyping makes the large-scale and moment-by-moment quantification of sensory phenotypes in psychiatric disorders highly scalable, it also introduces tremendous opportunities for genetic research and health improvement.

function of the human senses (seeing, hearing, smelling, tasting, and touching) responding to environmental stimuli. Accumulating evidence suggests that sensory defects represent the earliest symptoms of most psychiatric disorders [1][2][3][4]. As disease symptoms, sensory phenotypes may help clinical diagnosis and treatment through earlier detection of disease onset, relapse and improvement. Labeled as an endophenotype (having genetically-associated, predictable behavioral symptoms), sensory phenotypes may share some neurophysiological pathways common to psychiatric disorders; additionally, sensory phenotypes may also act as a direct index for neurophysiological effect [1]. Researchers have used sensory phenotypes to identify specific genotypes and to explain psychiatric disorders [5]. However, the existing low-throughput, high-cost methods for measuring sensory phenotypes are also impractical. They are not easily integrated into the massive data sets used in today's genomewide association studies (GWAS). Therefore, applying convenient and high-throughput digital technologies to measure sensory phenotypes carries great potential for studying psychiatric genetics.
Sensory phenotypes that measured with digital technologies are defined as digital sensory phenotypes. Using common digital technologies such as smartphones and wearable devices to measure and collect personal phenotypic data, termed "digital phenotyping" has tremendous potential [6][7][8]. Today, researchers are collecting phenotypic data from patients with psychiatric disorders including such parameters as daily mood, physical activities, and social communications using relatively inexpensive digital devices [9,10]. Digital phenotyping is unobtrusiveeven a normal smartphone can capture many types of phenotypic data. In psychiatry, objective and continuous quantitation of clinical markers using patients' own devices is useful to refine diagnosis, to tailor treatment strategy or monitor outcomes [11]. Through digital phenotyping, we can capture behavioral and sensor changes, and self-report information.
These changes should be distinguishable in nature and clinical status, and detectable by smartphone, wearable devices or other sensors. Among lots of human phenotypes, sensory phenotypes are eligible for the requirements.
In this review, we summarize studies that correlate sensory phenotypes to psychiatric disorders and those that use digital technologies to collect phenotypic data from patients with these psychiatric disorders.
We consider both sensory (perceptual) and sensorimotor functions as identifying sensory phenotypes. Sensorial functions refer to the basic abilities of sensory receptors and related neural circuitries, whereas sensorimotor functions refer to both sensory inputs and motor responses, e.g., eye movement or auditory EEG [12]. Abnormalities in sensory or sensorimotor functions suggest defects in the integrity of neural pathways and the nervous system as a whole [13]. Perception involves somewhat J Psychiatry Brain Sci. 2020;5:e200015. https://doi.org/10.20900/jpbs.20200015 Journal of Psychiatry and Brain Science 3 of 28 complex subjective judgment and is difficult to measure; therefore, we restricted our discussion in this essay to sensorimotor functions. Involving only limited little cognitive function, sensorimotor functions reflect sensory circuits directly. We are particularly interested in the feasibility of collecting sensory phenotypic data via digital technology and correlating those phenotypes to genetic variants associated with psychiatric disorders (see Figure 1).

Figure 1.
Overview: In this review, we introduce the concepts of the sensory phenotype and digital phenotyping for psychiatric studies. Furthermore, we discuss the feasibility of using digital technology to collect sensory phenotypes of psychiatric disorders.

PHENOTYPES IN PSYCHIATRIC GENETIC STUDIES
Diagnosis serves as a categorical phenotype for GWAS [1]; yet, diagnostic validity depends on how diagnosis is defined and whether its criteria are logically and factually reasonable [14][15][16]. For example, the clinical definition of autism has changed remarkably over the past 75 years. Once thought to be a form of childhood schizophrenia, autism is now considered a neurodevelopment disorder with genetic origins.
Diagnoses such as these are conventionally not based on expert observation and objective assessment and not on the physiological etiology [17,18]. Psychiatric disorders are difficult to classify due to their multidimensional phenotypes [16]. This difficulty is compounded by the burden of imprecise phenotyping, which impedes the identification of risk genes that contribute to psychiatric disease susceptibility [19].
The use of endophenotypes is one proposed method for linking disease diagnosis to genetic risk variant detection [20]. Endophenotypes are heritable, objective biological markers that can be measured directly.  [20]. Because they are directly measured and quantifiable, endophenotypes may be superior to traditional methods of diagnoses [20,21]. For example, sensory motorgating deficits consistently characterize schizophrenia [22][23][24]. Compared to complex disease behaviors, endophenotypes are governed by fewer genes. These genes may play an important role in the disease.
Endophenotypes may bridge disease diagnosis and gene identification, identifying "downstream" clinical phenotype traits as well as "up-stream" genetic output [20]. Endophenotypes may also help to identify aberrant genes in polygenic disease [25]. Furthermore, patients may be subclassified by specific endophenotypes [1]. Multiple endophenotypes could work together to constitute subtypes of the current diagnosis. The biology of endophenotypes contribute a fundamental understanding of the disease process, which has the potential to assist in prevention and more effective treatments [26].
Using endophenotypes offers a quantifiable method for diagnosis. This is plausibly a more precise and reproducible method than the qualitative, subjective categories of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). Furthermore, as a straightforward biological construct, endophenotypes are likely more accurate than conventional means in pinpointing a specific genetic abnormality and corresponding protein change [23]. Nonetheless, developing a truly high-grade and accurate endophenotype is a critical challenge. Researchers have identified and validated potential endophenotypes typically from disease-linked deficits [27]. Evidence of segregation and heritability in "clinically unaffected" relatives is a genotype-endophenotype correlation that commonly used for individual pedigree members [27]. For example, P50 suppression deficits were found as potential endophenotypes in patients with schizophrenia and unaffected relatives [28], which has been widely replicated and confirmed [26,[29][30][31][32][33]. Additionally, an association study identified the chromosomal region of interest [34], which yielded an association of P50 suppression deficits in schizophrenia via the α-7 subunit of the nicotinic receptor [35]. P50 suppression is considered to be one endophenotype of schizophrenia.
Several consortiums are attempting to apply multiple endophenotypes to large samples of patients [36]. Success will attest to the feasibility of endphenotypes for diagnosis within genetic studies [37]. phenotypic components of schizophrenia (SCZ) and bipolar disorder (BD) [38]. BSNIP measures physiological or cognitive traits, such as electroencephalography (EEG), eye movement activity and brain imaging.
Likewise, the Enhancing Neuroimaging and Genetics through Meta-Analysis (ENIGMA) Network uses advanced imaging technology to collect complex phenotyping data, identifying genetic influences on brain structure and function [39].

SENSORY PHENOTYPES IN PSYCHIATRIC DISORDERS
Sensory phenotypes are one of endophenotypes for psychiatric disorders. As the technology for collecting sensory measurements improves, interest among mental health professionals is growing.
Researchers can now capture previously undetectable sensory phenotypes and test various proposed phenotypes for their correlation to psychiatric characteristics. For example, deficits in the sense of smell are a sensory phenotype linked to negative symptoms in patients with schizophrenia [40]. One meta-analysis showed a deficient sense of smell in patients with schizophrenia and at-risk youth [41]. Wheras, patients with autism spectrum disorder (ASD) show heightened sensorimotor function.
ASD prompts atypical sensory reactivity and sensory over-responsivity [42,43], characterized by an extremely negative response to sensory stimuli [44]. ASD's sensory over-responsivity correlates to abnormal changes in the connectivity between the thalamus and the cortex [45].
Furthermore, eye-tracking deficiency in schizophrenia has also been investigated [46]. Chronic pain can be an example of a tactile phenotype of the whole body that is associated with anxiety and depression within epidemiological studies. Individuals with major depressive disorder (MDD) and other psychiatric disorders have an increased risk for chronic pain [47]. Recent research points to overlaps between pain-and depression-related changes in neuroplasticity and neurobiological mechanisms [48]. The sensory pathway of physical pain may involve multiple brain regions, including the insular and prefrontal cortices, associated with mood management [49].
To review the sensory phenotype study in psychiatric disorder, we In total, we selected 55 case-control studies, involving sensory J Psychiatry Brain Sci. 2020;5:e200015. https://doi.org/10.20900/jpbs.20200015 Journal of Psychiatry and Brain Science 6 of 28 phenotype of seven psychiatric disorders, including SCZ, BD, MDD, anxiety, OCD, AD, and HD. Significant differences between the disease and control populations were detected in case-control studies for several sensory phenotypes, e.g., eye-tracking in SCZ, BD, and MDD [50]; auditory eventrelated potential [51][52][53] in SCZ, BD, and anxiety; phenylthiocarbamide (PTC) non-taste in SCZ [54,55]; olfactory identification ability in SCZ [56] and AD [57]; and, tactile phenotype in OCD [58]. These studies focused primarily on differences in sensory phenotypes between patients with specific psychiatric disorders and healthy controls; they also reflect an interest in investigating the biological correlates of sensory function and psychiatric disorders. Tabulating the sample size and P-values of sensory phenotypic studies, researchers found that sensory phenotypes do significantly correlate with psychiatric disorders (Table 1). However, sample sizes were relatively small. In fact, most case-control studies had sample sizes under 400, with only two studies that used EEG measurements which included over 1000 participants. Sensory phenotypes are widely studied in the field of psychiatric disorders, however, mostly with small sample sizes.
In genetic studies, sensory phenotype research showed evidence of their correlation to psychiatric disorders. The fact that sensory phentopyes assist in detecting the genetic loci of psychiatric risks also attracts much attention [1]. One GWAS found that the sense of smell shares genetic regions with schizophrenia and Alzheimer's disease on chromosome 18 (see Table 2). [ [110] and practical, researchers are determined to improve its accuracy and convenience [111]. Likewise, measurement systems also are often time-consuming and inaccurate. The auditory event-related potential (ERP) is used to collect auditory phenotypes [52,53,64], measuring brain response related to sensory, cognitive, and motor events [112]. We found that the most frequently used measurement for hearing is the electroencephalogram (EEG) recording of the P50 and P300 waves [113]. increasingly own smartphones that could be used to benefit their health [116]. Researchers could collect patient's phenotypic data from smartphone sensors and wearable devices to determine health status [117].
Digital phenotyping encompasses the collection of data for symptoms relevant to psychiatric disorders as either passive or active data. "Passive data" refers to data produced with the patient's approval but without the patient having to iniatiate a response; these include GPS and accelerometer data collected by smartphones automatically [8]. Another digital phenotype analysis strategy is the ecological momentary assessment (EMA) [118,119] that uses "active data". "Active data" requires not only the patient's approval but also the patient's active involvement, such as taking surveys or contributing audio samples [8]. For instance.
For example, with the Beiwe smartphone platform [135], Barnett et al. [127] used mobility patterns and social behavior to predict relapse in schizophrenia. They found that the rate of behavioral anomalies was 71% higher within the two weeks preceding relapse. Saeb et al. [123] collected 48 college students' location sensor data and evaluated their depression symptoms severity using Patient Health Questionnaire 9-item (PHQ-9).

ANALYZE SENSORY PHENOTYPES
The range of available sensor input methods is wide and varied. Taps, clicks, scrolls, and cameras, with human-device interaction information provide multiple measures of sensory function. In regard to tactile phenotype, individuals with OCD, for example, can have abnormal touching patterns captured by a touchscreen [58,77]. For visual phenotypes, individuals with BD may possess inefficient eye-tracking and visual contrast sensitivity [50], which can be captured by a smartphone camera. Improved resolution and refresh frequency of phone cameras will allow the capture of increasingly complex traits related to eye movements that can currently only be analyzed using special devices [136,137]. With regard to auditory phenotypes, it is well known that individuals with BD and MDD can experience auditory verbal hallucinations in response to auditory stimuli. This can be measured by combining the capture of validated auditory stimuli through earphones and the user's interpretation documented on touchscreen [138]. Quality earphones and advanced audial technology can make studies related to sensitivity and the ability to differentiate direction and tone possible. In terms of smelling, tasting, and tactile phenotypes, individuals with SCZ often lose their ability to distinguish some smells and tastes [76]. While no sensors for measuring smell or taste-related phenotypes exist to date, patient-reporting of symptoms documented through smartphones could provide meaningful data.  [145,146]. Similar to electronic health records (EHR), digital technologies can also provide a longitudinal and comprehensive phenotypic record of sensory phenotype [147].
Privacy and data security are vital factors that must be considered with the use of digital sensory phenotyping [148,149] [155]. Similarly, the European Union implemented the General Data Protection Regulation, providing those nations with a legal framework to follow should data breaches occur [155].
This regulation might be a starting point for the development of similar regulatory processes for digital sensory phenotype data collection. For academia, it is essential to remember that collecting clinical data on human subjects requires adherence to globally accepted ethical regulations. They require scientific proof and the free volition of participants. Meanwhile, we should also seek ways to make such data generally available. Similar to John Sulston's advocacy that data from the Human Genome Project is openly accessible to the scientific community for common good, researchers should work to ensure that sensory phenotype data are used only for the common good [155]. Cricually, the ethical issues that face the realization of this technology may be more difficult to overcome than the technical hurdles. Although digital technology holds substantial potential for increasing access to mental healthcare, adequate solutions for safe data transmission and storage are needed to protect participant privacy. Establishing adequate protocols for data collection, data storage, and data process, as well as a framework for securing data usage is critical from the outset. Both the academic and government sectors must endeavor to ensure that data collection and analysis efforts are pursued equitably and transparently in the common interest of humankind. Governments should pass legislation restricting the use of data and protecting participants' privacy. While some researchers maintain a hopeful view of this new technology [157], its fruition relies upon advances in data security adequate to protect participants' privacy and to serve our common interests.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.