A Genome-Wide Association Study and Machine-Learning Algorithm Analysis on the Prediction of Facial Phenotypes by...

Area of Science:

• Genetics
• Dermatology
• Computational Biology

Background:

• Facial appearance is influenced by intrinsic and extrinsic factors, necessitating accurate personal skin condition assessment.
• Genetic identification of skin phenotypes is advancing through genome-wide association studies (GWAS) and machine learning.
• A need exists for large-scale GWAS in Asian populations due to a historical focus on European and American cohorts.

Purpose of the Study:

• To evaluate the correlation between facial phenotypes and candidate single-nucleotide polymorphisms (SNPs) in an Asian population.
• To develop a machine learning model for predicting facial phenotypes from genotype data.
• To identify novel SNPs associated with skin traits in Korean women.

Main Methods:

• Genome-wide association studies (GWAS) were conducted on 749 Korean women (aged 30-50) to identify SNPs associated with five facial phenotypes: melanin, gloss, hydration, wrinkle, and elasticity.
• Machine learning algorithms, including linear, ridge, and linear support vector regressions, were employed for phenotype prediction.
• Five-fold cross-validation was utilized to assess the performance of the prediction models.

Main Results:

• GWAS analysis identified 46 novel SNPs significantly associated with facial phenotypes (p < 1×10⁻⁵), including 3 for melanin, 20 for gloss, 12 for hydration, 6 for wrinkles, and 5 for elasticity.
• The ridge regression model demonstrated the highest prediction accuracy across all skin traits, with an R² ranging from 0.6422 to 0.7266.
• The study highlights the effectiveness of ridge regression for genotype-based skin phenotype prediction.

Conclusions:

• The developed facial phenotype prediction model, utilizing genotype information and machine learning, offers an optimal solution for accurate individual skin condition assessment.
• This approach has significant potential for the development of personalized cosmetics tailored to individual genetic profiles.
• The findings underscore the utility of integrating GWAS and machine learning for advancing dermatological insights and cosmetic applications in diverse populations.