Ensemble and Deep Learning Frameworks for Modeling Skin Phenotypes from Genetic and Lifestyle Inputs

Document Type

Conference Proceeding

Source of Publication

2026 14th International Conference on Bioinformatics and Computational Biology Icbcb 2026

Publication Date

3-1-2026

Abstract

Personalized dermatology requires integrating genetic predispositions with lifestyle factors to model individual skin health. This study consolidates three machine learning frameworks - multi-output gradient boosting, integrative clustering with Random Forest profiling, and a multi-task deep neural network (MT-DNN) - to advance precision dermatology through application-centric analysis.Using a multimodal dataset of 5,254 individuals (six skin-related genes and twenty-two lifestyle variables), the models jointly predict the severity of six skin phenotypes: Acne, Redness, Dryness, Sensitivity, Scarring, and Pigmentation. The multi-output LightGBM achieved near-expert agreement (average QWK ≥ 0.85), while unsupervised clustering revealed four dermatological subtypes that a Random Forest classifier predicted with 98% accuracy - surpassing single-modality models and emphasizing gene-lifestyle synergy. The MT-DNN captured nonlinear interactions such as AQP3 × stress and GPX1 × UV exposure, highlighting stress, sleep, and UV-related habits as dominant lifestyle drivers.Feature importance, SHAP, and statistical analyses ensured interpretability, translating complex dermatogenomic patterns into actionable insights. Overall, this work demonstrates that integrating complementary AI frameworks yields interpretable, high-accuracy modeling for personalized skincare, illustrating how existing algorithms can be repurposed to support bioinformatics-driven precision dermatology.

ISBN

[9798331545611]

Publisher

IEEE

First Page

173

Last Page

179

Disciplines

Computer Sciences | Medicine and Health Sciences

Keywords

Bioinformatics, Deep Neural Networks, Explainable AI, Gene-Lifestyle Interactions, LightGBM, Machine Learning, Multi-Output Learning, Multimodal Data Integration, Precision Dermatology, Random Forest, SHAP

Scopus ID

105046970931

Indexed in Scopus

yes

Open Access

no

Share

COinS