Using a Machine Learning Approach to Identify Low-Frequency and Rare FLG Alleles Associated with Remission of Atopic Dermatitis

Atopic dermatitis (AD) is a common relapsing inflammatory skin disease. FLG is the gene most consistently associated with AD. Loss-of-function variants in FLG have been previously associated with AD. Low-frequency and rare alleles (minor allele frequency < 5%) in this gene have been given less attention than loss-of-function variants. We fine sequenced the FLG gene in a cohort of individuals with AD. We developed a machine learning‒based algorithm to associate low-frequency and rare alleles with the disease. We then applied this algorithm to the FLG data, searching for associations between groups of low-frequency and rare FLG alleles and AD remission. A group of 46 rare and low-frequency FLG alleles was associated with increased AD remission (P = 2.76e-11). Overall, 16 of these 46 FLG variants were identified in an independent cohort and were associated with decreased AD incidence (P = 0.0007). This study presents an application of statistical methods in AD genetics and suggests that low-frequency and rare alleles may play a larger role in AD pathogenesis than previously appreciated.

Clinical Interpretation and Management of Genetic Variants

•

The human genome contains approximately 4 million variants, whose population frequencies vary according to the ethnic backgrounds.

•

Genetic diversity of humans in part determines interindividual variability in susceptibility to diseases, response to therapy, and the clinical outcomes.

•

Genetic variants exert a gradient of biological and clinical effect sizes. In general, variants with the largest effect sizes are responsible for the single-gene disorders, whereas those with moderate and modest effect sizes are responsible for oligogenic and polygenic diseases, respectively.

•

A phenotype is the consequence of nonlinear stochastic interactions among multiple genetic and nongenetic determinants.

•

Discerning pathogenicity of the genetic variants, identified through genetic testing, in the clinical phenotype is challenging and requires complementary expertise in human molecular genetics and clinical medicine.

Genetic variants are major determinants of susceptibility to disease, response to therapy, and clinical outcomes. Advances in the short-read sequencing technologies, despite some shortcomings, have enabled identification of the vast majority of the genetic variants in each genome. The major challenge is in identifying the pathogenic variants in cardiovascular diseases. The yield of the genetic testing has been limited because of technological shortcomings and our incomplete understanding of the genetic basis of cardiovascular disorders. To advance the field, a shift to long-read sequencing platforms is necessary. In addition, to discern the pathogenic variants, genetic diseases should be considered as a continuum and the genetic variants as probabilistic factors with a gradient of effect sizes. Moreover, disease-specific physician-scientists with expertise in the clinical medicine and molecular genetics are best equipped to discern functional and clinical significance of the genetic variants. The changes would be expected to enhance clinical utilities of the genetic discoveries.