Page 9 - Read Online

Pascolini. J Transl Genet Genom 2020;4:17-21. I https://doi.org/10.20517/jtgg.2020.05 Page 19

Figure 1. Schematic representation of the main phases of the present study and DeepGestalt technology experiment results. A: a
summary of the main phases of the study is provided; B: composite photos for the comparison analysis between the 4 study cohorts
(SETD5, Cohort 1; KBGS, Cohort 2; KdVS, Cohort 3, CTRL Ctrl.; unaffected controls, Cohort 4) (at the top). For each group, 18 2D frontal
images were analyzed. Confusion matrix (in the middle). The technology automatically detects facial points and recognizes dysmorphic
features through images analysis, evaluating similarities to known facial patterns (gestalt). Results of the SETD5 vs. KBGS and SETD5
vs. Ctrl. binary comparison experiment. Note the overlapping curves for the first comparison analysis, indicating a facial overlap of the 2
studied conditions (at the bottom). AUC: area under the curve; ROC: receiver operating characteristic

range between 0.66 and 0.98 [Figure 1B]. The mean accuracy was 78.47% with a standard deviation of
12.83%, and the random chance for comparison was 26.03% [Figure 1B]. Results of the binary comparison
experiment are expressed in terms of Area Under the Curve (AUC) and its Receiver Operating Characteristic
(ROC) curve for aggregated splits. Values result to be comprised between 0.735 (P = 0.050) and 1.000 (P =
0.000) (all experimental results are available in Supplementary Materials).

DISCUSSION
This is the first reported work to study the facial phenotype of the SETD5-associated ID, a very rare
malformation condition that is mainly characterized by neurodevelopmental delay, variable congenital
defects and dysmorphic craniofacial features. It can be included in the genetically heterogeneous group of
chromatin disorders, which represents a specific set of ID syndromes. These are mainly caused by mutations
in the various components of the chromatin remodeling BAF-complex (which are typically accompanied by
[5]
hypertrichosis as often happens in ID ) or in the histone’s modifiers, including SETD5. The latter encodes a
ubiquitously expressed methyltransferase containing a conserved domain of 130 amino acids and 2 signature
motifs (ELxF/YDY and NHS/CxxPN), acting as essential gene for normal embryo development and survival,
[6-9]
as has been demonstrated by previous in vivo studies .
[1]
Dysmorphisms are frequently encountered in patients with SETD5 mutations, as previously reported . To
better define the related phenotype, the facial appearance of affected individuals was compared, for the first-
time, with two other conditions with strong SETD5-ID clinical overlap and caused by mutations in molecules
involved in histones modification. Specifically, KBG (KBGS) and Koolen-de Vries (KDVS) syndromes, which
[11]
[10]
are respectively associated with mutations in ANKRD11 and KANSL1 , were considered. Both are clinically
defined by ID/DD, malformations and distinctive dysmorphisms, strongly resembling the SETD5-ID. This was
also confirmed by the CLINIC application of the Face2Gene platform, which identified KBGS (18/18, 100%) as
the most probable clinical diagnosis in patients with SETD5 variant, based exclusively on facial features. KdVS
was also proposed by the system for 4/18 (22%) individuals. KBGS is characterized by microcephaly, a round-