The diagnosis of rare genetic diseases is one of the most difficult areas in medicine. Whole-exome sequencing (WES) technology makes it easier to diagnose these diseases. In addition, next-generation phenotyping can help to diagnose computer-based algorithms. Detailed dysmorphologic findings of 25 patients diagnosed by WES in our center were described. The success of this technology in diagnosing rare genetic diseases was investigated by scanning the photographs of 25 patients with Face2Gene application. The application listed possible preliminary diagnoses (30 disease suggestion). Of these, 12 (48%) cases were correctly matched. The most common disease group in the patients was neurological disease (96%). The most common mode of inheritance in the patients was autosomal recessive. The rate of consanguineous marriages was determined in 80% of the patients. Ten patients had microcephaly and 7 patients had corpus callosum anomaly. In our study, we found that the success of Face2Gene was lower than described in the literature. We think that the probable cause of this condition is that the cases are very rare, and there is not enough data about these diseases in the application. Therefore, it is recommended that applications should be used more frequently by pediatricians and clinical geneticists. The diagnosis of rare diseases still is quite difficult. Nowadays, WES is a successful method. However, applications such as Face2Gene help to make a clinical prediagnosis and create a larger database.
Diseases with genetic etiology affect most of the population during their lifetimes. Syndromic cases have symptoms that significantly reduce the quality of life of patients, and they affect about 8% of the population [Baird et al., 1988]. Genetic diagnosis is important in syndromic cases because, following a diagnosis, there are special prevention and screening programs available for primary and secondary symptoms.
Many of these syndromic cases have facial phenotypes and some characteristic facial features that provide a clue for the diagnosis of genetic diseases. Approximately 30-40% of the genetic diseases show craniofacial alteration (as in Down syndrome, Fragile X, etc.) [Ferry et al., 2014].
Dysmorphology provides identification of before- or after-birth nonnormative forms that enables classification of various congenital malformations. This term provides a comparable description of all body characteristics (stature, feet, hands, neck) and face (such as shape of head, nose length, position of ears, thickness of vermillion, etc.) of the individuals of the same age group and ethnicity. A genetic etiology should be suspected if a child has a dysmorphic appearance with one of these features: (a) congenital anomalies, (b) growth retardation, (c) developmental delay and intellectual disability or developmental regression, (d) undeveloped secondary sexual characteristics, or (e) ambiguous genitalia [Featherstone et al., 2005; Smigiel and Demkow, 2016].
In the last 3 decades, with the advancement of technology, various databases have been created to determine the dysmorphic facial findings of patients (London Dysmorphology Database, Pictures Of Standard Syndromes and Undiagnosed Malformations Database, etc.) [Guest et al., 1999; Strømme, 1999]. These databases allow the comparison between the facial gestalt of patients and normal facial features, but these findings are affected by external factors such as available lighting conditions, posture of patients, etc. [Hammond, 2007]. In recent years, DeepGestalt program has been developed using computer technology and deep learning algorithms that measure similarities to hundreds of genetic syndromes based on unconstrained 2D images [Gurovich et al., 2018]. In a previous study, DeepGestalt achieved 91% top-10 correctness in identifying the syndrome on 502 different patient photos [Gurovich et al., 2019]. The Face2Gene (FDNA Inc, Boston, USA) application is a novel framework based on DeepGestalt, and it is one of the next-generation phenotyping (NGP) technologies used to determine the phenotype in thousands of diseases and correlate this phenotype with the genotype. In addition to DeepGestalt, which is an analysis based on a frontal facial photo of the patient, Face2Gene also offers an analysis based on the clinical findings of the patient (feature match). In this study, we considered both types of analysis and included them.
With the development of technology, other methods have been developed for the diagnosis of genetic diseases using molecular genetic tests. One of these methods is whole-exome sequencing (WES). This technique allows sequencing of the whole protein-coding region of genes in a genome (known as the exome) [Ng et al., 2009]. Exomes are about 2% of the whole genome, and mutations in these regions cause approximately 80% of mendelian diseases [Botstein and Risch, 2003; Yang et al., 2014]. In previous meta-analysis studies, the reported success rate of WES for molecular diagnosis was 24-68% (approximately 31%) [Clark et al., 2018]. This method has become a good alternative for the molecular diagnosis of rare genetic diseases that have prevalence below 1/2,000 [Ng et al., 2010].
Dysmorphic facial features are important to determine the causative variants of rare genetic diseases in WES. Therefore, when determining that pathogenic variants are causative of a disease or not, phenotypic features of patients are compared with similar previously reported cases using WES. For this reason, any type of clue (face gestalt, any prenatal findings, biochemical results, etc.) is increasingly important. In this study, we aimed to investigate the importance of determining dysmorphic facial features in rare genetic diseases with definitive molecular diagnosis using the WES method. We also evaluated the success of NGP technologies in this disease group.
Materials and Methods
The aim of this study was to determine the dysmorphic facial features of rare diseases diagnosed using WES and the power of NGP technology. This is a retrospective study. Therefore, the hospital registry system and patient files were reevaluated. The criteria for inclusion in the study are as follows: (1) patients were admitted to the Department of Medical Genetics at Afyonkarahisar University of Health Sciences between 2012-2019, (2) all patients had had sequencing results, (3) all patients had at least 1 front face image in the their files to evaluate the facial features, and (4) all patients had signed a consent form. Patients without adequate photographs to assess facial features were excluded from the study.
We selected 25 patients (9 females and 16 males) younger than 16 years of age who met all inclusion criteria. The photos and relevant clinical features were uploaded to Face2Gene, and the resulting analysis was assessed and correlated to the molecular diagnosis. We also checked how the correct diagnosis is ranked by both types of analysis, DeepGestalt and feature match. In addition, the research application of Face2Gene was used to understand whether the tool differentiated the group of patients from control groups [Pantel et al., 2018].
We used the hospital registry system to evaluate if the selected patients had anamnesis information, biochemical test results, MR images, echocardiographic results, and genetic reports. The parameters questioned in the anamnesis information were prenatal (oligohydroamnios, polyhydroamniosis, fetal akinesia, any abnormal ultrasound image), natal (delivery type, birth weight, gestation week at birth), postnatal stories (story of neonatal intensive care, hypoxic ischemic birth history, hypotonia, sucking difficulty), pedigree, presence of individuals with similar symptoms in the family, if consanguineous marriage (when yes, which degree), delay in intellectual/motor development stages, surgery history, and seizures.
Human Malformation Terminology (The Elements of Morphology) was used to determine the presence of any dysmorphic finding. This terminology was created by a group of clinicians working in the field of dysmorphology to standardize the definition of human morphology in which a reliable comparison of phenotypic findings among patients is provided [Allanson et al., 2009]. This terminology consists of 6 articles. These are standard terminology for the head and face; the periorbital region; the ear; the nose and philtrum; the lips, mouth, and oral region as well as the hands and feet.
WES analysis of the patients was performed by the contracted institutions or the universities we worked with in this study, and the test results were examined extensively. If the same gene was found to cause 2 or more diseases, differential diagnosis was made by detailed physical examination. Dysmorphic findings had a very important role at this stage. MR images and echocardiography findings were also used.
WES analysis findings of patients, anamnesis information, pedigree analysis, and clinical findings were collected. Dysmorphic facial features were evaluated in detail. Since dysmorphic facial features are a diagnostic clue to medical genetic doctors, we aimed to present dysmorphic facial findings of patients diagnosed as rare genetic diseases in our clinic. In this study, we evaluated the power of Face2Gene, a new NGP technology that makes recommendations for possible genetic disease using face gestalt information. Therefore, when choosing which test to perform in the diagnosis of rare genetic disease, we also evaluated the necessity of the recommendations of this next-generation approach.
The disease-causing mutations of all patients and their demographics are presented in Table 1. According to the mutation points, physical examination findings, laboratory and imaging tests, results were reevaluated and a definitive diagnosis was determined. Except for 1 patient, all of the diagnosed diseases were of autosomal recessive inheritance. In terms of consanguinity, consanguineous marriages were found in 80% (12 of them were first-degree cousins, 6 of them were second-degree, and 2 of them were third-degree cousins). No consanguinity was defined for parents of the other 5 patients, but they were from the same village (except for the parents of P25).
The symptoms and dysmorphic facial features in our patients are summarized in Table 2. Almost all of the patients were evaluated as neurological diseases (only 1 patient had a storage disorder). All other patients had neuromotor growth retardation and different degrees of learning disability. Ten patients (40%) had microcephaly. Two patients had muscle weakness, and 3 patients had persistent hypotonia. Only 1 patient had inadequacy in cerebellar tests. Twenty-two of the 25 patients had an MRI. When MRI findings were examined, 4 patients had normal findings, and 18 patients had abnormal findings. The most common findings were corpus callosum anomalies (7/18). Other anomalies were periventricular hyperintensity (4/18), dilated ventricles (4/18), and cortical atrophy (3/18), respectively.
Clinical findings of all patients were annotated in Face2Gene application. The most common clinical finding was global developmental delay (16%). The second most common findings were inability to walk, delayed speech and language development, hypospadias, hypoplasia of corpus callosum, and short stature (12% for all). The third most common findings were hypothyroidism, hyperreflexia and facial hypertrichosis (8% for all), and the fourth most common findings were hyperactive patellar reflex, hydrocephalus, hepatomegaly, postnatal growth retardation, equinovarus deformity and epiphyseal dysplasia (4% for all). The distribution of clinical features is shown in Figure 1.
Among the 30 diseases, the syndromes recommended by Face2Gene were analyzed for the presence and sequence of the syndrome diagnosed. Twelve of the cases (48%) had a correct match. The remaining 13 cases had a diagnosis that was not part of the 300 syndromes that DeepGestalt currently identifies. Thus, we only relied on the feature match algorithm. The application ranks the suggested syndromes based on 2 scores: Gestalt score and Feature score. Gestalt score is a value obtained by analyzing the photograph of the patient (dysmorphic facial features). The Feature score is a value obtained by analyzing the annotated clinical findings [Gurovich et al., 2019]. Gestalt score was calculated for only one (P9) of the patients included in this study. The other 11 patients had a Gestalt score as 0. Therefore, the Feature score is important for these patients. This situation shows the importance of the entry of the clinical feature findings of the patients to the application.
In this study, we presented the importance of dysmorphic findings in rare genetic diseases and the power of the DeepGestalt program, which enables facial analysis for genetic syndrome classification. This application helps clinicians to make a definitive diagnosis of the disease. With further development of this application, it may be possible to use it routinely in genetic polyclinics.
Prevalence of rare genetic diseases is between 6 and 8% [Lodato and Kaplan, 2013]. In this group of diseases, any clinical (prenatal, natal, and postnatal) and dysmorphic findings are of importance at the stage of prediagnosis. In this study, we presented the clinical, radiological, and dysmorphic features of patients with rare genetic diseases diagnosed by WES analyses. Regarding that clinical geneticists and pediatricians rarely encounter this group of diseases, collecting this information on a platform such as Face2Gene would be beneficial for the attending physicians. Because these applications are based on deep learning, the more patients analyzed, the higher the accuracy. Especially in rare genetic diseases, it is important to increase the Gestalt score after the analysis of the patient.
Gurovich et al.  reported that the success of Face2Gene top-10 matches was 91%. In 2019, Mishima et al. reported that Face2Gene success rate was 85.7% in patients with congenital dysmorphic syndromes in Japan. In the same study, if patients had a diagnosis for which Face2Gene had not been trained, the success of Face2Gene was 60.0% [Mishima et al., 2019]. We found low success of Face2Gene in consanguineous marriages. In our study, the recognition rate of Face2Gene application in our patient group was 48% (This rate was calculated from all diseases in a suggested list of 30 diseases). The probable reason for this low rate may be that the diseases in our patient group are not sufficiently introduced to Face2Gene; the rare genetic diseases are seen uncommonly in the population, and therefore have only been defined with a few patient photographs in this application.
The Face2Gene application was developed based on facial gestalt analysis. As a result of the face analysis, it determines the Gestalt score for 30 diseases matching the facial features of the patient. Furthermore, this application determines the Feature score for the 30 diseases by analyzing the anamnesis, clinical, laboratory and radiological findings of the patients. We strongly recommend that the clinical findings of the patient should be entered when using Face2Gene application, especially in rare genetic diseases. As seen in Table 3, data of patients with rare genetic diseases are limited in the application. We therefore recommend that physicians register the facial gestalt of patients to Face2Gene during or after diagnosis.
The diagnosis of rare genetic diseases is one of the most challenging fields in clinical genetics. The success rate of WES has been reported to be between 25% and 57% in the diagnosis of rare genetic diseases [Boycott et al., 2013; Stark et al., 2016]. However, in this patient group, it may be more helpful to analyze with NGP before, e.g., the DeepLearning program, and then follow traditional diagnostic methods considering the result of NGP. When considering the cost difference between targeted gene panels and WES, with this strategy, a diagnosis with lower costs can be provided [van Nimwegen et al., 2016], and at the same time, we believe the diagnostic procedures can be shortened.
Rare genetic diseases are often inherited in an autosomal recessive manner [Boycott et al., 2013]. Consanguinity between parents is one of the important risk factors for autosomal recessive inherited diseases. In the study of Hamamy et al. , it was reported that the prevalence of congenital anomaly in first-degree cousin marriage offspring increased by 1.7-2.8% compared to the general population. The rate of consanguineous marriages in Turkey as Muslim and Middle Eastern countries is high. In the study by Kelmemi et al. , the rate of consanguineous marriages in autosomal recessive inherited diseases was reported as 58% in Tunisia. In our study, 24 of 25 patients had autosomal recessive inherited disease; the rate of consanguineous marriages between the parents of patients with autosomal recessive inheritance was calculated as 80%. This higher rate may be due to the frequent consanguineous marriages in the region. In the study by Kelmemi et al. , the rate of parents sharing the same geographical origin in the autosomal recessive-inherited nonconsanguineous patient group was reported to be 63%.
Microcephaly is one of the most common clinical findings seen by medical genetic doctors. Whensearching for “microcephaly” in OMIM, the result was 1,007 from 8,976 entries. On the other hand, von der Hagen et al.  found that 28.5% of the children with microcephaly had genetic etiology. Also, in a study by Ashwal et al. , 15.5-53.3% of children with microcephaly had a genetic cause. In our study, the rate of patients with microcephaly who have a genetic disease was 40%. Symptoms such as microcephaly and dysmorphic facial features are common in genetic diseases. Therefore, online databases (such as OMIM) of clinical findings of diseases are helpful tools for clinical genetics. However, medical genetic doctors must decide which of the possible diagnoses is most plausible in the patient. This judgment requires extensive clinical experience.
Evaluating genetic diseases is like working as a detective who investigates a criminal event: every clue is very beneficial for reaching the diagnosis. One of these clues is MRI. Considering that most patients who consulted the genetic clinic are patients with neurological diseases, the importance of MRI is evident [Srivastava et al., 2014]. Corpus callosum abnormality is one of these MR findings. In a review article by Edwards et al. , 30-45% of the cases with corpus callosum agenesis were identified as genetic causes. In 25-30%, the reason is a single gene mutation. In a study by Schell-Apacik et al. , 32% of the patients with agenesis of the corpus callosum and dysgenesis of the corpus callosum had a genetic etiology. In our study, corpus callosum abnormality was found in 47% of the patients with a single gene mutation. The abnormal corpus callosum is a condition that should be taken into consideration by physicians because it may accompany a syndrome.
Nowadays, good anamnesis, examination, radiological and laboratory tests are necessary when diagnosing rare diseases. In addition, recently developed artificial intelligence technology can give us valuable clues in terms of diagnosis. The NGP technology may have a say in the diagnosis of genetic diseases with the use and contribution of people working in this field. So, we recommend employing such NGP programs.
In conclusion, each syndrome has a mask. In other words, all syndromes leave a trace or a clue in the phenotype. For years, dysmorphologists have been looking at photographs in dysmorphology books to diagnose patients' syndromes. With the widespread use of NGP technology in recent years, the diagnosis has become easier, not only by looking, but also by scanning patient pictures. However, introducing clinical findings to the program is crucial for more accurate diagnostic recommendations. Finally, with the expansion of such NGP programs, a large and powerful data library will be created, making future diagnostic procedures easier.
We thank Nicole Fleischer for her support and assistance.
Statement of Ethics
The project was approved by the University of Afyonkarahisar Health Sciences Ethics Committee, which covered the participating hospitals. All procedures followed were in accordance with the University of Sydney Human Research Ethics Committee and with the Helsinki Declaration of 1975, as revised in 2000 (5). Signed or electronic consent was obtained from all participants in the study.
The authors have no conflicts of interest to declare.
M. Elmas conceived and designed the study. B. Gogus performed clinical assessments, experiments, and contributed to data acquisition, analysis and interpretation. B. Gogus drafted the manuscript. Both authors contributed to critical revision of the manuscript for intellectual content and final approval of the manuscript.