 This short video describes a paper appearing in the current special issue of human mutation that focuses on deep phenotyping. The paper written by me, Peter Hammond and Mike Sutty describes techniques for large scale phenotyping of 3D facial morphology. These techniques are very visual in nature, so it seems best to show them in action rather than to describe them in words, and we begin by showing you 3D photographs of our own faces to my left and to my right. Both of these 3D photographs that you can see contain more than 25,000 surface points, which is plenty to capture even very subtle facial features. Laser scanners and CT and MRI devices can also be used for face shape analysis, but 3D photogrammetry is still the most commonly used modality in image acquisition. 3D facial phenotyping often begins by calculating and comparing average faces of homogeneous groups, and they could be healthy controls or individuals with the same micro-deletion or mutation. For example, this is the average face of a group of male individuals with Williams syndrome, which involves a micro-deletion of some 30th-O genes of the short arm of Chromazone 7. By contrast, the average face of a set of healthy male controls on the left clearly demonstrates differences in overall size and differences in shape in the nasal and perioral regions. These differences can be more clearly seen if we animate them in this morph between the average faces of the two groups. Now you can see more subtle differences such as the backward rotation of the mandrel and the fullness around the eyes. But lighting effects can mislead the appreciation of facial contours, so we consider shape-only comparisons without appearance, and these two, of course, can be animated. A more quantitative depiction of the difference is a heat map showing green regions of agreement with the control mean, and red regions of contraction and blue regions of expansion here is shown at two standard deviations, so the predominance of red emphasises that the growth delay is part of the Williams syndrome phenotype. Such comparisons of average faces of syndromic groups and controls are particularly useful for teaching dysmorphology to pediatricians and clinical geneticists. But if the interest is in phenotype-genotype comparison, then analyses need to be at a more individual level. These heat maps of the faces of 48 female individuals with Williams syndrome are normalised against healthy controls matched for age, sex and ethnic background. The regions of maximal red-blue intensity once again denote differences of two standard deviations, and we refer to each heat map as the face signature of the individual. Generally speaking, the more red and blue in the signature, the greater is the facial dysmorphism. A crude overall estimate of facial dysmorphism across all the 25,000 points on the face surface is the signature weight. It can be used to compare facial dysmorphism in groups. Here is the distribution of signature weight for the healthy female controls. As expected, the facial dysmorphism associated with Williams syndrome causes the associated signature weight to be significantly further to the right. Face signature can also be used to identify individuals whose face shape is atypical even for a dysmorphic condition like Williams syndrome. This scatter plot shows the face signature weight of female individuals that we looked at earlier. Horizontally, we normalise their facial dysmorphism against healthy controls and vertically against individuals with Williams syndrome. So 42 of these individuals form a reasonably homogeneous group, while 6 others are outliers. Compared to controls of similar age and sex, these two individuals have exceptionally small faces, hence the red in the heat map. Compared to others with Williams syndrome, their facial dysmorphism is coloured at a milder level. They are still at the edge, but within the pack. By comparison, these 4 individuals when compared to controls are at the mild end, but compared to others with Williams syndrome, they are extreme outliers. Indeed, the largely blue here indicates that their faces are unusually large for Williams syndrome. So as outliers, these 6 individuals may be of interest in genotype, phenotype studies. These signatures provide a panorama of facial dysmorphism, as in these 48 female individuals with Williams syndrome. If we inspect these just by eye, we can see similarities and differences, but we can induce more formally clusters of individuals with a similar pattern of facial dysmorphism. For example, these individuals obviously have a similar signature, and hence similar dysmorphism. We can arrange them to reflect who's dysmorphism is closest to whom, and we can link each individual to the other individual who's got the closest facial dysmorphism. If we repeat this for all 48 individuals, we get the associated safe signature graph. Individuals at the centre of this graph tend to be the least dysmorphic, while those at the peripherally are typically the most dysmorphic. That of course includes the 6 that we just found by looking at signature weight. Signature graphs can become much harder to interpret when we start to mix large groups of controls and affected individuals, as in this case. But by colour coding the nodes, we can distinguish between the controls, the empty circles, and affected individuals the filled circles. This coloured map form of the signature graph highlights this affected hypercluster, or collection of smaller clusters, suggesting they have a more homogeneous phenotype than the rest, who are more dispersed and hence less dissimilar to controls. So I hope that this brief description of facial phenotyping intriguing as wedges your appetite for reading the favourite fall, and I look forward to your comments and queries at the following email address.