Page 8 - Read Online

Page 18 Pascolini. J Transl Genet Genom 2020;4:17-21. I https://doi.org/10.20517/jtgg.2020.05

INTRODUCTION
Heterozygous mutations in the SETD5 gene (SET domain-containing protein 5, MIM#615743; KIAA1757)
have been associated with a syndromic intellectual disability (ID)/developmental delay (DD), also termed ID
[1]
autosomal dominant 23 (MIM#615761) .

The SET domain-containing protein 5 gene, mapped to the 3p25.3 genomic locus, encodes a histone-lysine
N-methyltransferase, which is involved in chromatin remodeling and is strongly expressed in both the adult
brain, as well as the spinal cord. Molecular bases of the disorder have also been recently demonstrated to be
[2]
associated with nonsense-mediated decay, resulting in gene haploinsufficiency .

The SETD5-related phenotype is mainly characterized by ID/DD, language delay and dysmorphic features.
To date, few patients (mostly pediatric) have been reported, being a very rare condition and initial
phenotype delineation has been provided by previous authors. The main craniofacial features are represented
by brachycephaly, long and smooth philtrums, micrognathia, synophrys, abnormal eyebrow, upslanted
palpebral fissures, bulbous nose with depressed nasal bridge and anteverted nares, thin upper lip vermilion,
downturned corners of the mouth, and dental crowding.

In this study, the SETD5-facial phenotype was analyzed with the DeepGestalt technology (V.19.1.3) (FDNA
Inc., Boston, MA, USA; https://www.face2gene.com) for the first time, to further define the craniofacial
characteristics, which should be considered as a key feature for clinical diagnosis.

METHODS
“SETD5” and “3p25.3 deletion/haploinsufficiency” terms have been digitized in the PubMed database
(https://www.ncbi.nlm.nih.gov/pmc/) to select all scientific works related to the SETD5-ID syndrome.
Deletions comprising not only SETD5 were included, based on evidence that it is the main gene determining
[2]
the typical core phenotype, as previously postulated . Only papers containing patients’ photographs were
considered for the experiment (for references details see Supplementary Materials). A total of 18 facial two-
dimensional (2D) images of individuals with SETD5 mutations were uploaded to the CLINIC and then
RESEARCH applications of the Face2Gene suite (SETD5, Cohort 1) [Figure 1A]. All patient images were
anonymized and uploaded to the personal account of the user (the author), which is protected by a password
and made inaccessible to others. All photos were processed by the system to generate unrecognizable,
composite matrices, which were then used for the comparison study (see below). The DeepGestalt
technology is based on deep learning algorithms built for syndrome-specific computational-based classifiers
(syndrome gestalts), converting patient photos into de-identified mathematical facial descriptors. The
patient’s facial descriptors are then compared to syndrome gestalts to quantify similarity (gestalt scores)
resulting in a prioritized list of syndromes with similar morphology. To date, available individuals with
SETD5 mutations are mostly pediatric (1-14 years) and Caucasian. Only one patient of Asian origin was
[3]
included in the present analysis . Cohort 1 was compared with three other cohorts: (1) one containing 18
published images of patients with KBG syndrome (KBGS, MIM#148050) (KBGS, Cohort 2); (2) 18 images of
patients with Koolen-de Vries syndrome (KdVS, MIM#610443) (KdVS, Cohort 3); and (3) the last consisting
of 18 images of unaffected individuals (Ctrl., controls, Cohort 4). KBGS and KdVS were included in the
study because both are chromatin disorders, which are caused respectively by mutations in the chromatin
modulator ANKRD11 (Ankyrin repeat domain-containing protein 11, MIM#611192) and KANSL1 (KAT8
regulatory NSL complex subunit 1, MIM#612452). Both conditions share strongly overlapping facial
features, including nasal and mouth dysmorphisms with the SETD5-related ID syndrome. The DeepGestalt
[4]
technology based on the Face2Gene platform was performed, as described previously .
RESULTS
Multiclass comparison analysis generated a confusion matrix, in which errors (false positives and false
negatives) were represented [Figure 1B]. The highlighted diagonal line indicates true positives values, which