GestaltMatcher Database - A global reference for facial phenotypic variability in rare human diseases

Area of Science:

• Genetics
• Medical Imaging
• Bioinformatics

Background:

• Facial dysmorphism presents significant challenges in diagnosing rare genetic disorders due to phenotypic variability.
• Next-Generation Phenotyping (NGP) tools struggle with diverse patient populations, limiting their clinical utility.
• Genetic ancestry influences facial features, complicating automated pattern recognition in medical imaging.

Purpose of the Study:

• To systematically analyze the impact of genetic ancestry on facial dysmorphism.
• To establish and utilize the GestaltMatcher Database (GMDB) for diverse rare disease facial imaging data.
• To evaluate the effect of data diversity on NGP tool performance.

Main Methods:

• Collected 10,980 frontal facial images from 8,346 patients with 581 rare disorders globally.
• Established the GestaltMatcher Database (GMDB) adhering to FAIR principles.
• Analyzed NGP performance using diverse training and testing datasets stratified by genetic ancestry.

Main Results:

• Incorporating non-European patient data significantly enhanced GestaltMatcher performance, increasing top-5 accuracy by +11.29%.
• Performance improvements were achieved without compromising accuracy for European patient data.
• GestaltMatcher Database (GMDB) now includes substantial data from Asian (19%) and African (7%) populations, with over 40% of African data being previously unpublished.

Conclusions:

• Cross-ancestral phenotypic variability in facial dysmorphism confounds NGP tools.
• International collaboration to increase data diversity is crucial for advancing NGP in clinical practice.
• The GestaltMatcher Database (GMDB) serves as a vital reference and training resource for clinicians and NGP development.