Prediction of skin color, tanning and freckling from DNA in Polish population: linear regression, random forest and neural network approaches

Abstract

Predicting phenotypes from DNA has recently become extensively studied field in forensic research and is referred to as Forensic DNA Phenotyping. Systems based on single nucleotide polymorphisms for accurate prediction of iris, hair and skin color in global population, independent of bio-geographical ancestry, have recently been introduced. Here, we analyzed 14 SNPs for distinct skin pigmentation traits in a homogeneous cohort of 222 Polish subjects. We compared three different algorithms: General Linear Model based on logistic regression, Random Forest and Neural Network in 18 developed prediction models. We demonstrate Random Forest to be the most accurate algorithm for 3- and 4-category estimations (total of 58.3% correct calls for skin color prediction, 47.2% for tanning prediction, 50% for freckling prediction). Binomial Logistic Regression was the best approach in 2-category estimations (total of 69.4% correct calls, AUC = 0.673 for tanning prediction; total of 52.8% correct calls, AUC = 0.537 for freckling prediction). Our study confirms the association of rs12913832 ( HERC2 ) with all three skin pigmentation traits, but also variants associated solely with certain pigmentation traits, namely rs6058017 and rs4911414 ( ASIP ) with skin sensitivity to sun and tanning abilities, rs12203592 ( IRF4 ) with freckling and rs4778241 and rs4778138 ( OCA2 ) with skin color and tanning. Finally, we assessed significant differences in allele frequencies in comparison with CEU data and our study provides a starting point for the development of prediction models for homogeneous populations with less internal differentiation than in the global predictive testing. Electronic supplementary material The online version of this article (10.1007/s00439-019-02012-w) contains supplementary material, which is available to authorized users.

Prediction of Eye, Hair and Skin Color in Admixed Populations of Latin America

“…A broad array of statistical methods have been employed in the literature to predict pigmentation traits, such as (multiple) linear [17,47] or (multinomial) logistic regression [10,11], decision trees [17,48], neural networks [17,49], and naïve Bayes classifiers [33,42,50]. Each method has its advantages and disadvantages, and are better suited for certain types of traits, e.g.…”

“…Each method has its advantages and disadvantages, and are better suited for certain types of traits, e.g. linear regression for quantitative traits [17,47] and logistic regression for categorical traits [10,11].…”

“…The pigmentation SNPs for prediction have been selected from GWAS performed mainly in Northern Europeans [4,5,17] and prediction models also optimized mostly in European datasets [17][18][19][20][21]. Furthermore, evaluation of the accuracy of these prediction tools has also been performed essentially in European-derived population samples [18][19][20]22].…”

Applying Random Forest Model Algorithm to GFR estimation

“…Many papers compared linear regression with random forest algorithms in other elds. As we can see, the random forest algorithm has not performed better in all eld and aspects (26)(27)(28), this reveals that random forest method can only take advantage over linear regression in some data models. So a more signi cant number of multi-centre data are needed to validate our outcomes in the eld of GFR estimation.…”

Interpol review of forensic biology and forensic DNA typing 2016-2019

“…The HIrisPlex-S system for eye, hair, and skin color prediction has been subjected to developmental validation studies [ 152 ]. Additional research studies have examined the effect of gender on eye color prediction [ 153 ], the ability to predict eye and hair color from World War II skeletal remains [ 154 ], the impact of age-depending hair color darkening during childhood [ 155 ], the prediction of head hair shape from DNA [ 156 ], the performance of four models for eye color prediction in an Italian population sample [ 157 ], and the development of new prediction models for skin color, tanning, and freckling from DNA in Polish populations using linear regression, random forest, and neural network approaches [ 158 ]. The predictability of tall stature from DNA markers has been explored in European samples [ 159 ] and genome-wide association studies conducted to identify loci influencing eyebrow color variation [ 160 ].…”