PEDIA: prioritization of exome data by image analysis

Abstract

Purpose: Phenotype information is crucial for the interpretation of genomic variants. So far it has only been accessible for bioinformatics workflows after encoding into clinical terms by expert dysmorphologists.

Methods: Here, we introduce an approach driven by artificial intelligence that uses portrait photographs for the interpretation of clinical exome data. We measured the value added by computer-assisted image analysis to the diagnostic yield on a cohort consisting of 679 individuals with 105 different monogenic disorders. For each case in the cohort we compiled frontal photos, clinical features, and the disease-causing variants, and simulated multiple exomes of different ethnic backgrounds.

Results: The additional use of similarity scores from computer-assisted analysis of frontal photos improved the top 1 accuracy rate by more than 20-89% and the top 10 accuracy rate by more than 5-99% for the disease-causing gene.

Conclusion: Image analysis by deep-learning algorithms can be used to quantify the phenotypic similarity (PP4 criterion of the American College of Medical Genetics and Genomics guidelines) and to advance the performance of bioinformatics pipelines for exome analysis.

Keywords: computer vision; deep learning; dysmorphology; exome diagnostics; variant prioritization.

Fig. 1

Prioritization of exome data by image analysis (PEDIA): cohort and classification approach. (a) Clinical features, facial photograph, and pathogenic variant of one individual of the PEDIA cohort. In total the cohort consists of 679 cases with monogenic disorders that are suitable for a diagnostic workup by exome sequencing. (b) Clinical features, images, and exome variants were evaluated separately and integrated to a single score by a machine learning approach. The disease-causing gene is shown at the top of the list.

Fig. 2

Performance readout and visualization of test results for a representative prioritization of exome data by image analysis (PEDIA) case. (a) For each case the exome variants are ordered according to four different scoring approaches, solely by a molecular deleteriousness score (CADD), by a score from image analysis (DeepGestalt), by a combination of a molecular deleteriousness score and a clinical feature–based semantic similarity score (CADD+Phenomizer), or the PEDIA score that includes all three levels of evidence. The sensitivity of the prioritization approach depends on the number of genes that are considered in an ordered list. The top 1 and top 10 accuracy rates correspond to the intersection of the curves at maximum rank 1 and 10. Note that for benchmarking DeepGestalt on the gene level, syndrome similarity scores first have to be mapped to the gene level, resulting in a lower performance compared with the readout on a phenotype level, due to heterogeneity. The area under the curve is largest for PEDIA scoring. (b) The disease-causing gene of the case depicted in Fig. 1 achieves the highest PEDIA score and molecularly confirms the diagnosis of Coffin–Siris syndrome. Other genes associated with similar phenotypes, such as Nicolaides–Baraitser syndrome, also achieved high scores for gestalt but not for variant deleteriousness.