Dermainformatics: A Data-Driven Framework for Function-Based Skin Typing and Causal Estimation of Skin Health Outcomes

Dermainformatics: A Data-Driven Framework for Function-Based Skin Typing and Causal Estimation of Skin Health Outcomes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Dermainformatics: A Data-Driven Framework for Function-Based Skin Typing and Causal Estimation of Skin Health Outcomes Rie Nakamura, Kohei Kanno, Yoshichika Noda, Akihiro Tanaka, Chikako Yoshikawa, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8013431/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Conventional skin type classifications rely on subjective, phenotype-based descriptors (e.g., dryness, oiliness, sensitivity), limiting mechanistic insight and predictive value. We introduce Dermainformatics, novel data-driven framework designed to separate biological skin function from its clinical phenotype and embed causal inference. Thirteen biophysical/biochemical indicators (e.g., stratum corneum hydration, transepidermal water loss (TEWL), desmoglein-1 (DSG1), sebum and fatty-acid composition, collagen content, skin thickness, skin temperature) were clustered using k-means with silhouette-based model selection, yielding three functional profiles (N = 165). We then estimated marginal (population-average) effects of profiles on expert-graded clinical and appearance-related binary outcomes and image-derived continuous measures using stabilized inverse probability of treatment weighting (IPTW) with prespecified confounders (age, sex, skincare habit, current smoking). Primary analyses applied global 1st/99th-percentile trimming of stabilized IPTW (Option A); robustness was evaluated with covariate-balance diagnostics (|SMD|<0.1), weight distributions and effective sample size, and a winsorization sensitivity at the same percentiles. Profiles exhibited distinct functional signatures with coherent, population-average associations for wrinkles, redness, transparency, acne, pores, and porphyrin features (reported as odds ratios and standardized β). By decoupling function from phenotype and quantifying marginal causal effects within the region of covariate overlap, Dermainformatics yields interpretable, mechanism-aligned stratification suitable for personalized skincare and population-level risk prediction, and is readily extensible to multi-omics and lifestyle data for precision prevention. This framework provides a reproducible, physiology-grounded basis for quantitative skin typing and population-level risk estimation. As a generalizable contribution to computational dermatology, Dermainformatics facilitates preventive and personalized skincare by coupling function-based profiling with causal inference. Health sciences/Biomarkers Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Health care Health sciences/Medical research computational dermatology skin informatics causal inference inverse probability weighting risk stratification personalized skincare Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Skin type classification is a foundational construct in dermatology and cosmetic science, guiding skincare recommendations and therapeutic choices. However, these prevailing schemes are built on subjective surface-level descriptors that fail to fully capture the vast physiological heterogeneity of skin function and offer limited predictive power for future conditions. Representative examples include the Fitzpatrick phototype, which is anchored in the UV response [1, 2], and the Baumann Skin Type (BSTI), which stratifies individuals into 16 categories via a questionnaire [3, 4] however, these descriptive systems provide limited mechanistic specificity “which functional pathways are involved” and limited actionability “what to change to modify risk”. Recent progress in standardized, non-invasive measurements has enabled multidimensional quantification of skin function. Barriers and hydration can be assessed by transepidermal water loss (TEWL) and stratum corneum hydration under the European Group on Efficacy Measurement of Cosmetics and Other Topical Products (EEMCO) guidance[5–7]. Cell adhesion/keratinization markers such as desmoglein-1 (DSG1) implicate epidermal cohesion and barrier integrity [8]. Sebum and lipid composition (fatty-acid profiles) relate to the microbiome, inflammation, and acne pathophysiology [9, 10]. Dermal structure, including skin thickness and collagen content, can be quantified using high-frequency ultrasound [11, 12], and skin temperature serves as a physiological functional marker linked to thermoregulation and barrier behavior [13]. Thus, ostensibly similar phenotypes (e.g., “oily” vs “dry”) may arise from distinct biological pathways—including barrier disruption, lipid-metabolic imbalance and inflammation, or structural degradation. Therefore, we first isolated and summarized the functional layer (functional profiling) and then quantified its causal downstream impact on phenotypes to move from descriptive labeling toward mechanism-aligned, actionable stratification. Any observational study is inherently vulnerable to confounding factors. Accordingly, rigorous adjustment at the design stage is essential to derive estimates that carry a meaningful interventional interpretation. We adopt marginal structural models (MSMs) with stabilized inverse probability of treatment weighting (IPTW) to target marginal (population average) effects[14, 15]. Implemented as a weighting design, MSM/IPTW simplifies the outcome model and is comparatively robust to functional-form misspecification. It also addresses the non-collapsibility of the odds ratio in logistic models, whereby conditional effects do not need equal marginal effects, even without confounding [16, 17]. Moreover, settings with three functional profiles (multilevel exposure) can leverage multinomial propensity scores (e.g., multinomial logistic regression or generalized boosted models) for weight estimation [18]. Against this background, we developed Dermainformatics (Fig. 1 ). This novel framework takes a "function-first" approach, beginning with the separate measurement of biological function and clinical phenotype and then rigorously connecting them through a causal lens. Specifically, we (1) defined three functional profiles via unsupervised clustering of 13 indicators; (2) estimated stabilized IPTW using prespecified confounders (age, sex, skincare habit, current smoking); and (3) quantified marginal causal effects on expert-graded binary outcomes and image-derived continuous outcomes. Primary analyses used global 1st/99th-percentile trimming (Option A) to restrict inference to the region of covariate overlap, and validity was evaluated via covariate balance (|SMD|<0.1), weight distributions, and effective sample size (ESS), with winsorization at the same percentiles as a sensitivity analysis. This design clarifies how functional biology maps phenotypic risks, supporting both personalized care and population-level prediction. a, Inputs: 13 functional indicators and phenotypic outcomes. b, Unsupervised functional profiling by k-means (k = 3 selected by silhouette) with descriptive visualization. c, Multinomial propensity scores and stabilized IPTW; primary handling of extreme weights by global 1st/99th-percentile trimming (Option A); winsorization at the same cutoffs used only for sensitivity. d, Diagnostics: covariate balance (max |SMD|≤0.1), weight distribution and weighted ESS, positivity via PS overlap, and trimming-vs-winsorization agreement (Deming slope, Lin’s CCC). e, Weighted marginal structural models yield population-average effects (OR for binary; standardized β for continuous). Methods Study design and participants This cross-sectional observational study enrolled healthy Japanese participants aged 18–39 years at the Hokkaido Information University between May 2024 and June 2024. Participants were recruited from a volunteer panel through recruitment notices. All measurements were obtained under standardized indoor conditions (temperature, 21–24°C; relative humidity, 40–60%). The analyses used complete cases for 13 functional indicators (N = 165). For downstream outcomes, primary analyses used global 1st/99th-percentile trimming of stabilized IPTW (Option A), so denominators reflect non-missing observations after weighting and trimming. The inclusion and exclusion criteria and device models are detailed in the Supplementary Methods. Skin function measurements We assessed 13 biophysical and biochemical indicators in the cheek to create a comprehensive functional signature of the skin. This spanned four key domains: barrier/hydration, lipid biology, dermal structure, and thermal physiology. Biophysical: transepidermal water loss (TEWL), water-holding capacity (hydration capacity), stratum corneum hydration, dermal hydration, homogeneity (spatial uniformity), sebum amount, skin thickness, collagen content, and skin temperature. Biochemical (tape stripping): desmoglein-1 (DSG1) protein abundance[19], palmitoleic acid (C16:1), oleic acid (C18:1), and total free fatty acids (FFAs)[20]. All the devices were operated by trained staff according to the manufacturer’s recommendations under stable environmental conditions. Prior to the analysis, all indicators were z-standardized (mean 0, s.d. 1). Unsupervised clustering for functional profiling To derive functional skin profiles, we applied k-means clustering to 13 standardized indicators. The models for k = 2–6 were fitted with k-means + + initialization, n_init = 100, and a fixed random seed. The average silhouette width served as the primary model selection criterion; the Calinski–Harabasz and Davies–Bouldin indices and cluster sizes were summarized as secondary diagnostics. The selected number of clusters and diagnostic values are reported in the Results and Supplementary material. The resulting profiles (Profiles 0–2) served as exposure in the causal analyses. Outcomes We analyzed two outcome families: Binary outcomes (expert visual grading): wrinkle, turnover, transparency, seborrheic dermatitis, rosacea, redness, and acne (collapsed to binary per prespecified rules). Continuous outcomes (image-derived): Cheek and forehead domains spanning wrinkles, brown spots, red spots, pores, and textures, plus skin color parameters L*, a*, and b*. To report standardized coefficients, each outcome was standardized using the weighted mean and variance of the analytic sample. Detailed outcome definitions, image processing, and quality control are provided in the Supplementary Methods. Propensity score and stabilized IPTW To adjust for prespecified confounding, we estimated multinomial propensity scores \(\:{e}_{a}\left(X\right)=\text{Pr}\left(A=a\:\right|X)\) for the three-level exposure \(\:A\in\:\{\text{0,1},2\}\) using multinomial logistic regression with age, sex (female=1), skincare habit (yes=1), and current smoking (yes=1) as covariates. The stabilized inverse probability of treatment weights (IPTW) was computed as $$\:{w}_{i}=\frac{\text{P}\text{r}(A={A}_{i})}{{e}_{{A}_{i}}\left({X}_{i}\right)}$$ Primary handling of extreme weights (Option A): We trimmed observations with stabilized IPTW outside the global 1st/99th percentiles and applied this uniformly across outcomes (primary analysis). The realized pooled thresholds are 0.4368 and 2.6360 respectively. Covariate balance was evaluated using absolute standardized mean differences (|SMD|). For a three-level exposure, pairwise |SMD| was summarized as the maximum |SMD| per covariate. Weight distributions (histogram/KDE) and the weighted effective sample size (ESS) after trimming were summarized to characterize the weighting design. Outcome models and estimands Our estimate was the marginal (population average) effect in the region of common support. We fit weighted marginal structural models (MSMs) with outcome profile indicators (profile 0 as a reference). Binary outcomes: weighted logistic MSMs; effects are reported as odds ratios (OR) with 95% confidence intervals (CI) for Profile 1 vs. 0 and Profile 2 vs. 0. Continuous outcomes: weighted linear MSMs; effects are reported as standardized coefficients (β) with 95% confidence intervals CI. Standardization used the weighted outcome s.d. as the denominator. Multiple testing correction was applied to functional profiling (Kruskal–Wallis test with BH-FDR). For downstream outcomes, we prioritized estimation across correlated endpoints, reporting effect sizes, and 95% CIs, rather than significance screening; no across-the-board multiplicity adjustment was applied. Domainwise adjusted summaries can be provided upon request. Sensitivity analysis We applied 1st/99th-percentile winsorization to the stabilized IPTW at the same thresholds as the primary trimming (0.4368/2.6360), without exclusions, and re-fitted all models to assess robustness to the handling of extreme weights. Agreement between the primary trimming and winsorization estimates was quantified using Deming error-in-variables regression (λ = 1; accounts for error on both axes) and Lin’s concordance correlation coefficient (CCC) together with mean/median absolute errors (MAE/MedAE). For binary outcomes, we compared log odds ratios; for continuous outcomes, we compared the standardized β. We also summarized the proportion of pairwise differences within a region of practical equivalence (ROPE) (± 0.10 for log-OR; ±0.05 for β). High CCC and Deming slopes near 1 were pre-specified as evidence that weight handling does not materially distort direction or magnitude. Missing data Clustering and propensity score estimation used complete cases for the 13 indicators and confounders (N = 165). Outcome-specific analyses used non-missing observations after weighting and trimming; denominators therefore vary by outcome and are reported in Figures/tables (see Supplementary Table S4 for counts and ESS). Software Analyses were performed in Python (3.11) using pandas, scikit-learn (clustering), statsmodels (weighted MSMs, robust SE), and matplotlib for visualization. Random seeds were fixed to enhance the reproducibility. Code and exact package versions are listed in the code availability statement. Ethics The study was approved by Hokkaido Information University (approval number: 2023-14). All procedures adhered to the Declaration of Helsinki and written informed consent was obtained from all participants. Results This section describes the three functional skin profiles derived from unsupervised clustering. It then confirmed the validity of the causal inference methodology before presenting the marginal effects of these profiles on a range of binary and continuous clinical phenotypes. 3.1 Functional profiling and between-group differences Thirteen skin-function indicators (hydration, transepidermal water loss [TEWL], lipid-related measures, structural markers, and temperature) were z-standardized and clustered using k-means (k = 2–6; k-means++; n_init = 100; fixed seed). The average silhouette width peaked at k = 3 (0.127) and slightly decreased at k = 4 (0.123), whereas the Calinski–Harabasz index decreased (31.49→27.51) and the Davies–Bouldin index improved (2.204→1.906) from k = 3 to k = 4 (Supplementary Fig. S1). Therefore, we retained k = 3 for the primary analyses. The cluster size was 70/56/39 (N = 165). The unweighted baseline characteristics (All and Profiles 0–2) are summarized in Supplementary Table S1. Functional characteristics were visualized using radar plots (Fig. 2a), PCA (Fig. 2b), Cohen’s d heat map (Fig. 2c), and indicator ranking by max |d| (Fig. 2d). Formal comparisons used Kruskal–Wallis with Benjamini–Hochberg FDR; pairwise effect sizes (d for 0 vs. 1, 0 vs. 2, and 1 vs. 2) and max |d| ranks are provided in Supplementary Table S2. Because we separated function (putative causes) from phenotype (downstream outcomes), these differences are presented as descriptive contexts; causal estimates are reported as marginal effects below. Based on these functional signatures, we hereafter refer to Profile 2 as the "hyperseborrheic" profile, Profile 1 as the " barrier-impaired " profile, and Profile 0 as the "balanced" reference profile. 3.2 Covariate balance and weight diagnostics (stabilized IPTW) Multinomial propensity scores were estimated for the three-level exposure (Profiles 0–2) using prespecified covariates (age, sex, skincare habit, and current smoking). Stabilized IPTW achieved excellent covariate balance; in the Love plot (Fig. 3a), the maximum pairwise |SMD| per covariate fell within the 0.1 threshold post-weighting. The weight distribution (Fig. 3b) was centered near 1.0; primary analyses applied global 1st/99th-percentile trimming (Option A) with realized pooled thresholds of 0.4368 and 2.6360. Detailed weight summaries are provided in Supplementary Tables S3a (trim) and S3b (winsorization sensitivity). The outcome-specific denominators and weighted effective sample sizes (ESS) after trimming are summarized in Supplementary Table S4, together with weight statistics. 3.3 Binary phenotypes: clinical interpretation (stabilized IPTW with Option A) Fig. 4 shows the marginal odds ratios from weighted logistic MSMs after stabilized IPTW with primary trimming (Option A). The clinical findings were consistent. Profile 2: Age-related changes come to the fore, with consistently elevated wrinkle risk. Inflammatory/seborrheic features, such as redness, acne, and seborrheic dermatitis, also tended to be higher. Profile 1: Loss of transparency and slowed turnover dominate, and seborrheic dermatitis is also higher, consistent with dryness, low-grade inflammation, and barrier compromise. Profile 0: binary endpoints for inflammation and ageing remain comparatively low. Trends were visualized via a forest plot (Fig. 4a), log(OR) heat map (Fig. 4b), and profile-wise rankings (Fig. 4c–d). Full estimates are provided in the supplementary Table5. 3.4 Continuous phenotypes: focused domains of texture and tone (stabilized IPTW with Option A) Fig. 5 reports standardized coefficients (β) from weighted linear MSMs for five representative domains—wrinkle, pigmentation, pore, texture, and red spot; the remaining continuous outcomes are summarized in Supplementary Table S6. The clinical read-out is: Profile 2: Worsening of pore counts and texture roughness was most pronounced, followed by deterioration of wrinkles and pigmentation. These texture-related changes tend to move in the same direction across both the cheek and forehead. Profile 1: Pigmentation (spots) tends to increase, with red spots (mottled erythema) and texture deterioration standing out; the primary signal is a dulling of overall skin tone. Profile 0: these domains remain comparatively favorable. The direction and relative magnitude are conveyed by forest plots (Fig. 5a, a′) and heat maps (Fig. 5b, b′), and the rankings (Fig. 5c–d, c′–d′) indicate which measures most strongly characterize each profile. 3.5 Sensitivity analysis: robustness to extreme weights (Option A) Estimates from primary trimming closely tracked those from winsorization across both binary (log-OR) and continuous (standardized β) outcomes (Fig. 6). Deming slopes were close to unity with small intercepts, Lin’s CCC indicated excellent concordance, and absolute errors were modest. The majority of pairwise differences fell within the ROPE, supporting the insensitivity of the conclusions to the specific weight-handling rule. Detailed agreement metrics and side-by-side estimates are provided in Supplementary Materials. Discussion This study introduces Dermainformatics, a function-first, causality-oriented framework that separates skin function from phenotype, and quantifies the population-average (marginal) effects of functional profiles on clinically salient outcomes. By deriving profiles from objective biophysical and biochemical indicators—rather than from visible attributes— we ground skin typing in measurable physiology and then connected that physiology to phenotypic risk using stabilized IPTW. A central contribution is the directional separation of predictors (functions) from outcomes (phenotypes). Profiles were defined solely from 13 functional indicators (hydration, barrier loss via TEWL, lipid composition and sebum, structural markers, and temperature) and only thereafter linked to clinical endpoints. This design respects causal temporality, reduces outcome-dependent circularity in classification, and enables the estimation of meaningful interventional marginal effects under standard assumptions (consistency, positivity, and no unmeasured confounding). The resulting profiles were mapped to distinct clinical images. Profile 2 (hyperseborrheic) shows the clearest aging-linked signature, with wrinkle risk most consistently elevated and texture roughness and pore counts most pronounced [21], followed by worsening pigmentation, which tends to move in the same direction across the cheek and forehead. In parallel, inflammation/seborrheic features (redness, acne, and seborrheic dermatitis) also tended to be higher, consistent with sebaceous activity, microbial metabolites (e.g., porphyrins) [22, 23], and low-grade inflammation that can accelerate extrinsic aging pathways and collagen degradation [24]. Profile 1 (barrier-impaired) centers on loss of transparency and slowed turnover, with pigmentation (spots), red spots (mottled erythema), and texture deterioration, and the dominant clinical signal is dulling of the overall tone[25, 26]. These patterns align with the reduced hydration/barrier integrity and microinflammation. Profile 0 remained comparatively favorable across domains and served as a functional reference. Methodologically, Dermainformatics extends skin typing beyond description in two ways. First, weighting-based confounding control (stabilized IPTW) targets marginal effects that are directly actionable for prevention and population risk management; diagnostics confirmed a good balance (|SMD|<0.1), suitable weight distributions, and robustness to the treatment of extreme weights. In the primary analysis, we applied global 1st/99th-percentile trimming (Option A) to restrict inference to the region of covariate overlap; winsorization at the same percentiles (no exclusions) served as a sensitivity analysis, and the conclusions were concordant [15, 27, 28]. Second, the framework pairs biophysical metrics with specimen-derived biochemical markers (e.g., DSG1, fatty acid composition), capturing both the quantity and quality of barrier/sebaceous function, and improving the physiological interpretability of clusters[19, 29]. These findings have practical implications for personalized skincare and trial design. For Profile 2, programs prioritizing lipid regulation, antioxidant/anti-inflammatory strategies, and texture/pore management may mitigate both inflammatory burden[22] and aging-linked texture/wrinkle trajectories [30]. For Profile 1, barrier restoration and turnover support (e.g., humectants, barrier lipids, gentle keratolytics, and anti-inflammatory care) may address the core “dullness” phenotype while reducing erythema heterogeneity[25]. Because the effects are marginal and estimated within overlap, these recommendations translate naturally to population-level guidance and enrichment strategies in interventional studies. This study has some limitations. The design is observational and cross-sectional; causal interpretation relies on assumptions and measured confounders (age, sex, skincare habit, and smoking)[31, 32]. Generalizability is bounded by the study population (healthy Japanese participants aged 18–39 years) and by site-specific measurements and device choices. Unsupervised clustering entails modeling decisions (k selection, standardization) that we probed with diagnostics, but remain open to alternative formulations. Finally, while profiles are physiologically interpretable, they are not diagnostic labels; they summarize functional states that can be modified or co-occur. Future work should test stability across populations, seasons, and devices, assess longitudinal transitions between functional states, and embed interventional trials that target modifiable functions (barrier, lipids, and micro-inflammation) to verify risk reduction. Integrating multi-omics, microbiome, and digital lifestyle data will likely refine profiles and enhance prediction, while preserving the causal separation of function from phenotype. Conclusion Dermainformatics reframes skin typing as a mechanism-aligned functional profiling coupled with marginal causal effects on phenotypes. By separating the function from the outcome, weighting confounders with stabilized IPTW, and validating with diagnostics and sensitivity analyses, the framework yields interpretable, actionable risk estimates. Profile 2 (hyperseborrheic) was most associated with aging-related deterioration, with pronounced texture/pore changes and consistent wrinkle elevation, whereas Profile 1(barrier-impaired) was dominated by tone dullness (transparency loss, slowed turnover) alongside pigmentation and red-spot heterogeneity. These insights support profile-specific prevention and provide a scalable template for precision dermatology that can be extended using longitudinal and multimodal data. Dermainformatics reframes skin typing as a generalizable analytical framework in computational dermatology. By separating biological functions from phenotypes and coupling them with stabilized IPTW, it yields population-level, interpretable risk estimates that support risk stratification and personalized recommendation design. Rather than emphasizing immediate clinical deployment, our contribution is a reproducible, physiology-grounded basis for quantitative skin typing that can scale across cohorts and devices. Future work will extend to longitudinal validation, external cohorts, and multimodal integration (e.g., microbiome and lifestyle sensors) while preserving the causal separation of function from phenotype. Declarations Data availability The de-identified data that support the findings of this study are available from the corresponding author upon reasonable request for peer review, within the terms of participant consent and applicable data-use agreements. Upon acceptance, the final dataset will be deposited in an open repository and a persistent DOI will be provided in the published version. Code availability Reproducible analysis code (Python; exact package versions listed in Methods) will be provided to editors and reviewers via a private repository during peer review. Upon acceptance, the code and environment files will be released under an open-source license and archived with a persistent DOI. Code and minimal reproducible materials will be made publicly available upon acceptance. Acknowledgement This work was supported by the Cross-ministerial Strategic Innovation Promotion Program (SIP), “Building a sustainable food chain that provides abundant and nutritious food” (funding agency: Bio-oriented Technology Research Advancement Institution). Funding This study was primarily supported by KOSÉ Corporation. The Cross-ministerial Strategic Innovation Promotion Program (SIP), “Building a sustainable food chain that provides abundant and nutritious food(tentative)” (funding agency: Bio-oriented Technology Research Advancement Institution) provided partial support limited to data acquisition (instrument procurement and field logistics), as this research serves as a foundational step in a broader initiative to explore the link between nutritional status and skin health. Funders had no role in the design, analysis, interpretation, or publication decision. Author Contributions Statement N.R. conceived the study, designed the methodology, performed the data analysis, and wrote the original manuscript. N.R. is the lead and corresponding author. K.K. and Y.N. contributed to the study design, performed the on-site data acquisition, and curated the data. A.T. contributed to the study of design and data management. C.Y., K.S., and S.S. conducted the analysis of the samples collected. All authors have read and approved the final manuscript. Competing interests The authors declare no competing interests. References Eilers, S., et al., Accuracy of Self-report in Assessing Fitzpatrick Skin Phototypes I Through VI. JAMA Dermatology, 2013. 149 (11): p. 1289-1294. Sommers, M.S., et al., Are the Fitzpatrick Skin Phototypes Valid for Cancer Risk Assessment in a Racially and Ethnically Diverse Sample of Women? Ethn Dis, 2019. 29 (3): p. 505-512. Baumann, L., Understanding and treating various skin types: the Baumann Skin Type Indicator. Dermatol Clin, 2008. 26 (3): p. 359-73, vi. Lee, Y.B., et al., Which Skin Type Is Prevalent in Korean Post-Adolescent Acne Patients?: A Pilot Study Using the Baumann Skin Type Indicator. Ann Dermatol, 2017. 29 (6): p. 817-819. Berardesca, E., EEMCO guidance for the assessment of stratum corneum hydration: electrical methods. Skin Res Technol, 1997. 3 (2): p. 126-32. Berardesca, E., et al., The revised EEMCO guidance for the in vivo measurement of water in the skin. Skin Res Technol, 2018. 24 (3): p. 351-358. Rogiers, V., EEMCO guidance for the assessment of transepidermal water loss in cosmetic sciences. Skin Pharmacol Appl Skin Physiol, 2001. 14 (2): p. 117-28. Sherrill, J.D., et al., Desmoglein-1 regulates esophageal epithelial barrier function and immune responses in eosinophilic esophagitis. Mucosal Immunol, 2014. 7 (3): p. 718-29. Del Rosso, J.Q. and L. Kircik, The primary role of sebum in the pathophysiology of acne vulgaris and its therapeutic relevance in acne management. Journal of Dermatological Treatment, 2024. 35 (1): p. 2296855. Cao, K., et al., Fatty Acid Profiling in Facial Sebum and Erythrocytes From Adult Patients With Moderate Acne. Frontiers in Physiology, 2022. Volume 13 - 2022 . Czajkowska, J., et al., High-frequency ultrasound in anti-aging skin therapy monitoring. Scientific Reports, 2023. 13 (1): p. 17799. Meng, Y., et al., Application of high-frequency ultrasound to assess facial skin thickness in association with gender, age, and BMI in healthy adults. BMC Med Imaging, 2022. 22 (1): p. 113. MacRae, B.A., et al., Skin Temperature Measurement Using Contact Thermometry: A Systematic Review of Setup Variables and Their Effects on Measured Values. Frontiers in Physiology, 2018. Volume 9 - 2018 . Robins, J.M., M.Á. Hernán, and B. Brumback, Marginal Structural Models and Causal Inference in Epidemiology. Epidemiology, 2000. 11 (5). Hernán, M.A., B. Brumback, and J.M. Robins, Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology, 2000. 11 (5): p. 561-70. Greenland, S., Noncollapsibility, confounding, and sparse-data bias. Part 1: The oddities of odds. J Clin Epidemiol, 2021. 138 : p. 178-181. Greenland, S., Noncollapsibility, confounding, and sparse-data bias. Part 2: What should researchers make of persistent controversies about the odds ratio? Journal of Clinical Epidemiology, 2021. 139 : p. 264-268. McCaffrey, D.F., et al., A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med, 2013. 32 (19): p. 3388-414. Naoe, Y., et al., Bidimensional analysis of desmoglein 1 distribution on the outermost corneocytes provides the structural and functional information of the stratum corneum. J Dermatol Sci, 2010. 57 (3): p. 192-8. Camera, E., et al., Use of lipidomics to investigate sebum dysfunction in juvenile acne [S]. Journal of Lipid Research, 2016. 57 (6): p. 1051-1058. Roh, M., et al., Sebum output as a factor contributing to the size of facial pores. Br J Dermatol, 2006. 155 (5): p. 890-4. Spittaels, K.J., et al., Porphyrins produced by acneic Cutibacterium acnes strains activate the inflammasome by inducing K(+) leakage. iScience, 2021. 24 (6): p. 102575. Mayslich, C., P.A. Grange, and N. Dupin, Cutibacterium acnes as an Opportunistic Pathogen: An Update of Its Virulence-Associated Factors. Microorganisms, 2021. 9 (2). Shimizu, N., et al., Oxidation of squalene by singlet oxygen and free radicals results in different compositions of squalene monohydroperoxide isomers. Scientific Reports, 2018. 8 (1): p. 9116. Green, M., et al., Transepidermal water loss (TEWL): Environment and pollution-A systematic review. Skin Health Dis, 2022. 2 (2): p. e104. Galizia, G., A. Belloni Fortina, and A. Semenzato, Seborrheic Dermatitis: From Microbiome and Skin Barrier Involvement to Emerging Approaches in Dermocosmetic Treatment. Cosmetics, 2024. 11 (6): p. 208. Robins, J.M., M.A. Hernán, and B. Brumback, Marginal structural models and causal inference in epidemiology. Epidemiology, 2000. 11 (5): p. 550-60. Cole, S.R. and M.A. Hernán, Constructing Inverse Probability Weights for Marginal Structural Models. American Journal of Epidemiology, 2008. 168 (6): p. 656-664. Cohen-Barak, E., et al., The Role of Desmoglein 1 in Gap Junction Turnover Revealed through the Study of SAM Syndrome. J Invest Dermatol, 2020. 140 (3): p. 556-567.e9. Nakamura, R., et al., Prediction of future wrinkles for middle-aged women: A 7-year longitudinal study on the progression of wrinkles in Japanese women. Skin Research and Technology, 2021. 27 (5): p. 854-862. Austin, P.C. and E.A. Stuart, Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med, 2015. 34 (28): p. 3661-79. Austin, P.C., Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med, 2009. 28 (25): p. 3083-107. Additional Declarations No competing interests reported. Supplementary Files SupplementaryTableS1.xlsx SupplementaryTablesS2.xlsx SupplementaryTableS3a.xlsx SupplementaryTableS3b.xlsx SupplementaryTableS4.xlsx SupplementaryTableS5.xlsx SupplementaryTableS6.xlsx Supplemetarymaterialssummary.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8013431","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":549349815,"identity":"574e1f0f-83d9-4ab8-9654-de80b8dadd3b","order_by":0,"name":"Rie Nakamura","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvElEQVRIiWNgGAWjYFACxgYgYWMA4zITo6URqCeNJC1gaw4bEFQGB+YNzO0PPradN+bnP3yA4UcFA7s5IS0yBxgbG2e23TaTnJGWwNhzhoHZsoGAFgn5h43NvG23bQxu8BgwM7YxMBscIKQF6H2glnM2BufPfyBJywEzgwM5DMRrmTnjXLIx0C8GB3vOSBDhFwb2Bx8+lNkZ9vMffvjgR4VNMsEQQwFAJ0kkkxBDUGBHupZRMApGwSgY7gAA4e8457lmqMgAAAAASUVORK5CYII=","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":true,"prefix":"","firstName":"Rie","middleName":"","lastName":"Nakamura","suffix":""},{"id":549349817,"identity":"b50f3efe-2cad-4e9d-aab3-fa6705133330","order_by":1,"name":"Kohei Kanno","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Kohei","middleName":"","lastName":"Kanno","suffix":""},{"id":549349819,"identity":"a6cc39ae-c099-46ee-8935-71adf0b27112","order_by":2,"name":"Yoshichika Noda","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Yoshichika","middleName":"","lastName":"Noda","suffix":""},{"id":549349820,"identity":"98c48c4c-c01a-4659-89e5-8332dbf6c90e","order_by":3,"name":"Akihiro Tanaka","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Akihiro","middleName":"","lastName":"Tanaka","suffix":""},{"id":549349823,"identity":"8e9f2cb9-6cec-4b27-92a2-bf5124af8cee","order_by":4,"name":"Chikako Yoshikawa","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Chikako","middleName":"","lastName":"Yoshikawa","suffix":""},{"id":549349828,"identity":"02210a73-a0fa-4715-8135-c8d77e5c99ff","order_by":5,"name":"Ken Suematsu","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Ken","middleName":"","lastName":"Suematsu","suffix":""},{"id":549349830,"identity":"48a23565-cb8c-427c-9080-1931c3321859","order_by":6,"name":"Shunichi Suga","email":"","orcid":"","institution":"KOSÉ (Japan)","correspondingAuthor":false,"prefix":"","firstName":"Shunichi","middleName":"","lastName":"Suga","suffix":""}],"badges":[],"createdAt":"2025-11-02 23:08:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8013431/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8013431/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96787739,"identity":"726f6bbd-10ff-4e53-91a9-36ca15f80330","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":749202,"visible":true,"origin":"","legend":"","description":"","filename":"manuscriptSciRep.docx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/1384e05076df3859e959a897.docx"},{"id":96787735,"identity":"0429db11-780e-4264-8418-adb6d06d8a53","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"json","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9362,"visible":true,"origin":"","legend":"","description":"","filename":"300928f4929945cd917bfae3af4a67b5.json","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/67ec6b7bcd3ccfe5a5ffb4d2.json"},{"id":96787740,"identity":"f6adcd7b-c126-4f9b-a31b-de74c237b8ae","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9739,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/ec3da167f008fe7e0db46c64.xlsx"},{"id":96787750,"identity":"67f53490-600c-4e2e-832f-874f53bc4930","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10915,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS3a.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/a6a681ef66b3459611119531.xlsx"},{"id":96787746,"identity":"11927b80-3fcf-445f-ae53-c533cfefbf99","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10891,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS3b.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/36d2ed3172661b9965e74030.xlsx"},{"id":96787755,"identity":"acb6a181-c017-4c33-b340-93c9471c13f8","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11316,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/37f7fb7e40350584146d9a1b.xlsx"},{"id":96787753,"identity":"d014f18d-dc35-4d38-844d-4c4b2aadfd9a","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11873,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS5.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/cb79c8107ac117436ba5b42b.xlsx"},{"id":96787758,"identity":"6c75d416-22ab-44ba-a14b-1b6e57df2ac7","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13764,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS6.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/8ec9c3987777d842e3800843.xlsx"},{"id":96916946,"identity":"f68a71a4-c07e-4b5c-b7a8-23ff72316e50","added_by":"auto","created_at":"2025-11-27 14:09:05","extension":"xlsx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15568,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTablesS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/7efb4522430dbd96a60a1b2b.xlsx"},{"id":96787768,"identity":"063c59d6-b419-46b3-9c21-6ed8cd61a6ed","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"docx","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":261218,"visible":true,"origin":"","legend":"","description":"","filename":"Supplemetarymaterialssummary.docx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/c16576302200deca9165e81c.docx"},{"id":96787757,"identity":"0ccb0e93-139c-4c4b-aff9-5ab436b7c186","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":67590,"visible":true,"origin":"","legend":"","description":"","filename":"300928f4929945cd917bfae3af4a67b51enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/28dff6934630332ecc6950ee.xml"},{"id":96787762,"identity":"1c8504d6-7e8b-43e2-b8b1-ef7b38473eab","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"pptx","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2173854,"visible":true,"origin":"","legend":"","description":"","filename":"figssubmit.pptx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/fd1f70f9fe1219e7e7610253.pptx"},{"id":96916142,"identity":"8dec7d75-fcb9-494d-8105-06517dc1235a","added_by":"auto","created_at":"2025-11-27 14:08:05","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":203541,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/da5c85eb6b0fde8228b02977.png"},{"id":96787765,"identity":"3e701e64-2d36-4c78-8303-8594f23d62b7","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123100,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/1d34860737f02e7e396c2341.png"},{"id":96787766,"identity":"62b5a849-3d4b-4c57-af0f-af47b9b24351","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":68282,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/6c6511feda7780bb4d9dfea9.png"},{"id":96915916,"identity":"fc862915-8922-4f75-8ff9-5420c2a8ae3f","added_by":"auto","created_at":"2025-11-27 14:07:46","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":85126,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/9f6f9c70d07ab2dfe8edc700.png"},{"id":96787748,"identity":"667e0ecf-13bd-4f03-8e58-83b1153e21f7","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"jpeg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":279901,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/c12a36f1e8992faeb35d02a3.jpeg"},{"id":96915548,"identity":"2992849f-a72b-421e-8c5a-44d917c05505","added_by":"auto","created_at":"2025-11-27 14:07:22","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96761,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/82b49147275c96d593b63038.png"},{"id":96787761,"identity":"28229d26-6f66-45a3-b198-78eaec3ece54","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":71242,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/95849cd05cbeae2180148c73.png"},{"id":96787772,"identity":"1c37bf47-24b1-492e-bc10-bff9deb57ddf","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25125,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/21b6c9670661e85a0b05bda9.png"},{"id":96787769,"identity":"f96e4209-6353-4390-bb4d-a6e7d50ac8e0","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16828,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/00fed0c178fa4679e4db2a00.png"},{"id":96787747,"identity":"a2fb586c-0ec1-4221-bfcd-3701fe60b79a","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16698,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/96dbe1589307e8cbcd9563e9.png"},{"id":96787752,"identity":"3f60d83c-c59e-4bdf-a0af-fd71a17d9359","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43742,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/72974f89b07555f808d31e03.png"},{"id":96787773,"identity":"f360a5d0-b326-42d9-8eff-8cfc570ba48e","added_by":"auto","created_at":"2025-11-26 06:27:36","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40194,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/a8004760b3d2a5cfb7c6cb68.png"},{"id":96787763,"identity":"2a20367a-e7c6-4fa3-a804-bfc5f3465a6e","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xml","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65313,"visible":true,"origin":"","legend":"","description":"","filename":"300928f4929945cd917bfae3af4a67b51structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/754a6863d2765dff796c849c.xml"},{"id":96787770,"identity":"aa4065c7-10e6-4b9c-80d3-cbe4232b0576","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"html","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76483,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/71a219e9deed99214ac24dd3.html"},{"id":96787733,"identity":"ec133085-a015-42d6-be49-e8862b6ec423","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":653349,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of the Dermainformatics pipeline.\u003c/p\u003e\n\u003cp\u003ea, Inputs: 13 functional indicators and phenotypic outcomes. b, Unsupervised functional profiling by k-means (k=3 selected by silhouette) with descriptive visualization. c, Multinomial propensity scores and stabilized IPTW; primary handling of extreme weights by global 1st/99th-percentile trimming (Option A); winsorization at the same cutoffs used only for sensitivity. d, Diagnostics: covariate balance (max |SMD|≤0.1), weight distribution and weighted ESS, positivity via PS overlap, and trimming-vs-winsorization agreement (Deming slope, Lin’s CCC). e, Weighted marginal structural models yield population-average effects (OR for binary; standardized β for continuous).\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/cbae4070a4520ba33058ea12.png"},{"id":96787742,"identity":"a4084ffa-e6cd-477c-a552-4c995b3c05be","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":118901,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional characterization of skin profiles.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea, Radar plots of 13 indicators normalized to a 0–1 scale (profile means) for profiles 0–2 (n = 70/56/39; N = 165).\u003c/p\u003e\n\u003cp\u003eb, Principal component analysis (PCA) of z-standardized indicators; axes show explained variance (PC1, 29.0%; PC2, 14.3%).\u003c/p\u003e\n\u003cp\u003ec, Heat map of Cohen’s d for all pairwise profile contrasts per indicator (pooled s.d.; the sign follows the left–right order in the column labels).\u003c/p\u003e\n\u003cp\u003ed Indicators ranked by maximum absolute d; bar label indicates the profile pair attaining the maximum. The exact values are presented in Supplementary Table S2.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/b9bfad0a33bcea2668533a76.png"},{"id":96787741,"identity":"ad007578-54cc-48ff-905c-1dbe0a605d5d","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":86125,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional characterization of skin profiles.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea, Radar plots of 13 indicators normalized to a 0–1 scale (profile means) for profiles 0–2 (n = 70/56/39; N = 165).\u003c/p\u003e\n\u003cp\u003eb, Principal component analysis (PCA) of z-standardized indicators; axes show explained variance (PC1, 29.0%; PC2, 14.3%).\u003c/p\u003e\n\u003cp\u003ec, Heat map of Cohen’s d for all pairwise profile contrasts per indicator (pooled s.d.; the sign follows the left–right order in the column labels).\u003c/p\u003e\n\u003cp\u003ed Indicators ranked by maximum absolute d; bar label indicates the profile pair attaining the maximum. The exact values are presented in Supplementary Table S2.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/f47f65e253e45db3fe32735a.png"},{"id":96787744,"identity":"101d7c46-f378-4162-9950-b26587fb4793","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":133541,"visible":true,"origin":"","legend":"\u003cp\u003eCovariate balance and weight diagnostics after IPTW.\u003c/p\u003e\n\u003cp\u003ea, Love plot of absolute standardized mean differences (|SMD|) for prespecified confounders (age, sex, skincare habit, smoking) before and after stabilized IPTW. For three-level exposure, |SMD| was computed for each pairwise profile contrast and summarized as the maximum |SMD| per covariate. Dashed line: |SMD| \u0026lt; 0.1.\u003c/p\u003e\n\u003cp\u003eb,Histogram and kernel density estimate (KDE) of stabilized IPTW. Vertical lines mark the median (~1.0) and \u0026nbsp;global 1st/99th-percentile thresholds (0.4368/2.6360) used in the primary trimming analysis (Option A).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/b03091160c9ca55b58199b2f.png"},{"id":96787749,"identity":"78753528-9b69-47ea-8680-4591c28092ea","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":279901,"visible":true,"origin":"","legend":"\u003cp\u003eAssociations between functional skin profiles and binary phenotypic outcomes after stabilized IPTW (Option A).\u003c/p\u003e\n\u003cp\u003ea, Forest plot of odds ratios (OR) with 95% confidence intervals (CI) for seven binary outcomes (wrinkle, turnover, transparency, seborrheic dermatitis, rosacea, redness, and acne), comparing profile 1 vs. 0 and profile 2 vs. 0.\u003c/p\u003e\n\u003cp\u003eb, Heat map of log(OR) values showing the direction and magnitude of associations.\u003c/p\u003e\n\u003cp\u003ec–d Outcomes ranked by OR for Profile 1 vs. 0 (c) and Profile 2 vs. 0 (d). Stabilized IPTW was estimated from age, sex (female = 1), skincare habit (yes = 1), and current smoking (yes = 1), with global 1st/99th percentile trimming (0.4368–2.6360; Option A) in the primary analysis. The vertical dashed lines indicate null values (OR = 1). Sample sizes corresponded to non-missing observations after trimming\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/98791d9d857ce395d0f2bf0a.jpeg"},{"id":96916069,"identity":"b6140456-f236-488f-8034-99d77c8a4599","added_by":"auto","created_at":"2025-11-27 14:07:57","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":173643,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAgreement between trimming and winsorization of stabilized IPTW estimates.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePanels compare the effect estimates obtained with winsorization (x-axis) versus percentile trimming (y-axis) of stabilized inverse-probability weights. Dashed line: identity. Solid line: Deming regression (error-in-variables). Lin’s concordance correlation coefficient (CCC) and demand slope are shown in each panel.\u003c/p\u003e\n\u003cp\u003ea, Binary outcomes (log OR).\u003c/p\u003e\n\u003cp\u003eb, Continuous outcomes (standardized β).\u003c/p\u003e\n\u003cp\u003eAgreement is high in both families, indicating that the direction and magnitude of the effects are stable across weight handling choices.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/939b62c0a6ab6df9a9fa3fcd.png"},{"id":98435218,"identity":"55676142-6c41-4440-ab02-c50cab4c58e3","added_by":"auto","created_at":"2025-12-17 16:53:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2283740,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/9baea4e7-5f5c-46c6-b46c-579a5d7d3e28.pdf"},{"id":96915077,"identity":"53f49db9-c3ee-4885-b510-b3966194469a","added_by":"auto","created_at":"2025-11-27 14:06:50","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":9739,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/020fd6cedb42729241d4f02b.xlsx"},{"id":96915909,"identity":"5e75a176-c2d0-48ef-85dd-3011b10f552f","added_by":"auto","created_at":"2025-11-27 14:07:46","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":15568,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTablesS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/c45d6c4dea1a132b998860ae.xlsx"},{"id":96787737,"identity":"a478f26f-9b49-4ae0-951a-9a17fd73bc9a","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10915,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS3a.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/676902c1a5fb00d3a378a276.xlsx"},{"id":96916630,"identity":"d66a41ec-bc89-4988-b77a-977a3f34da01","added_by":"auto","created_at":"2025-11-27 14:08:47","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":10891,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS3b.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/e78875ef41148e838abbe9af.xlsx"},{"id":96787743,"identity":"b4082cc5-b0b4-42be-aafd-04f874d59b44","added_by":"auto","created_at":"2025-11-26 06:27:34","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":11316,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/7d38417c3996337e4836accb.xlsx"},{"id":96915346,"identity":"b38f42a8-3cc7-4fa5-8c1d-a7a0b0767c37","added_by":"auto","created_at":"2025-11-27 14:07:09","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":11873,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS5.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/fc37eee47b793ed3a1e46ddb.xlsx"},{"id":96787745,"identity":"faff4d70-4ec0-48c8-babe-71cc60fb7e17","added_by":"auto","created_at":"2025-11-26 06:27:35","extension":"xlsx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":13764,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS6.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/809854cbb00cc1c569134aef.xlsx"},{"id":96917645,"identity":"ebe774e4-57f4-4d71-8b9b-8c871dd9cc49","added_by":"auto","created_at":"2025-11-27 14:10:19","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":261218,"visible":true,"origin":"","legend":"","description":"","filename":"Supplemetarymaterialssummary.docx","url":"https://assets-eu.researchsquare.com/files/rs-8013431/v1/4443e83f9aa8d3283aa4b9b6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Dermainformatics: A Data-Driven Framework for Function-Based Skin Typing and Causal Estimation of Skin Health Outcomes","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSkin type classification is a foundational construct in dermatology and cosmetic science, guiding skincare recommendations and therapeutic choices. However, these prevailing schemes are built on subjective surface-level descriptors that fail to fully capture the vast physiological heterogeneity of skin function and offer limited predictive power for future conditions. Representative examples include the Fitzpatrick phototype, which is anchored in the UV response [1, 2], and the Baumann Skin Type (BSTI), which stratifies individuals into 16 categories via a questionnaire [3, 4] however, these descriptive systems provide limited mechanistic specificity \u0026ldquo;which functional pathways are involved\u0026rdquo; and limited actionability \u0026ldquo;what to change to modify risk\u0026rdquo;.\u003c/p\u003e\u003cp\u003eRecent progress in standardized, non-invasive measurements has enabled multidimensional quantification of skin function. Barriers and hydration can be assessed by transepidermal water loss (TEWL) and stratum corneum hydration under the European Group on Efficacy Measurement of Cosmetics and Other Topical Products (EEMCO) guidance[5\u0026ndash;7]. Cell adhesion/keratinization markers such as desmoglein-1 (DSG1) implicate epidermal cohesion and barrier integrity [8]. Sebum and lipid composition (fatty-acid profiles) relate to the microbiome, inflammation, and acne pathophysiology [9, 10]. Dermal structure, including skin thickness and collagen content, can be quantified using high-frequency ultrasound [11, 12], and skin temperature serves as a physiological functional marker linked to thermoregulation and barrier behavior [13]. Thus, ostensibly similar phenotypes (e.g., \u0026ldquo;oily\u0026rdquo; vs \u0026ldquo;dry\u0026rdquo;) may arise from distinct biological pathways\u0026mdash;including barrier disruption, lipid-metabolic imbalance and inflammation, or structural degradation. Therefore, we first isolated and summarized the functional layer (functional profiling) and then quantified its causal downstream impact on phenotypes to move from descriptive labeling toward mechanism-aligned, actionable stratification.\u003c/p\u003e\u003cp\u003eAny observational study is inherently vulnerable to confounding factors. Accordingly, rigorous adjustment at the design stage is essential to derive estimates that carry a meaningful interventional interpretation. We adopt marginal structural models (MSMs) with stabilized inverse probability of treatment weighting (IPTW) to target marginal (population average) effects[14, 15]. Implemented as a weighting design, MSM/IPTW simplifies the outcome model and is comparatively robust to functional-form misspecification. It also addresses the non-collapsibility of the odds ratio in logistic models, whereby conditional effects do not need equal marginal effects, even without confounding [16, 17]. Moreover, settings with three functional profiles (multilevel exposure) can leverage multinomial propensity scores (e.g., multinomial logistic regression or generalized boosted models) for weight estimation [18].\u003c/p\u003e\u003cp\u003eAgainst this background, we developed Dermainformatics (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). This novel framework takes a \"function-first\" approach, beginning with the separate measurement of biological function and clinical phenotype and then rigorously connecting them through a causal lens. Specifically, we (1) defined three functional profiles via unsupervised clustering of 13 indicators; (2) estimated stabilized IPTW using prespecified confounders (age, sex, skincare habit, current smoking); and (3) quantified marginal causal effects on expert-graded binary outcomes and image-derived continuous outcomes. Primary analyses used global 1st/99th-percentile trimming (Option A) to restrict inference to the region of covariate overlap, and validity was evaluated via covariate balance (|SMD|\u0026lt;0.1), weight distributions, and effective sample size (ESS), with winsorization at the same percentiles as a sensitivity analysis. This design clarifies how functional biology maps phenotypic risks, supporting both personalized care and population-level prediction.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003ea, Inputs: 13 functional indicators and phenotypic outcomes. b, Unsupervised functional profiling by k-means (k\u0026thinsp;=\u0026thinsp;3 selected by silhouette) with descriptive visualization. c, Multinomial propensity scores and stabilized IPTW; primary handling of extreme weights by global 1st/99th-percentile trimming (Option A); winsorization at the same cutoffs used only for sensitivity. d, Diagnostics: covariate balance (max |SMD|\u0026le;0.1), weight distribution and weighted ESS, positivity via PS overlap, and trimming-vs-winsorization agreement (Deming slope, Lin\u0026rsquo;s CCC). e, Weighted marginal structural models yield population-average effects (OR for binary; standardized β for continuous).\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy design and participants\u003c/h2\u003e\u003cp\u003eThis cross-sectional observational study enrolled healthy Japanese participants aged 18\u0026ndash;39 years at the Hokkaido Information University between May 2024 and June 2024. Participants were recruited from a volunteer panel through recruitment notices. All measurements were obtained under standardized indoor conditions (temperature, 21\u0026ndash;24\u0026deg;C; relative humidity, 40\u0026ndash;60%). The analyses used complete cases for 13 functional indicators (N\u0026thinsp;=\u0026thinsp;165). For downstream outcomes, primary analyses used global 1st/99th-percentile trimming of stabilized IPTW (Option A), so denominators reflect non-missing observations after weighting and trimming. The inclusion and exclusion criteria and device models are detailed in the Supplementary Methods.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eSkin function measurements\u003c/h3\u003e\n\u003cp\u003eWe assessed 13 biophysical and biochemical indicators in the cheek to create a comprehensive functional signature of the skin. This spanned four key domains: barrier/hydration, lipid biology, dermal structure, and thermal physiology.\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eBiophysical: transepidermal water loss (TEWL), water-holding capacity (hydration capacity), stratum corneum hydration, dermal hydration, homogeneity (spatial uniformity), sebum amount, skin thickness, collagen content, and skin temperature.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eBiochemical (tape stripping): desmoglein-1 (DSG1) protein abundance[19], palmitoleic acid (C16:1), oleic acid (C18:1), and total free fatty acids (FFAs)[20].\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eAll the devices were operated by trained staff according to the manufacturer\u0026rsquo;s recommendations under stable environmental conditions. Prior to the analysis, all indicators were z-standardized (mean 0, s.d. 1).\u003c/p\u003e\n\u003ch3\u003eUnsupervised clustering for functional profiling\u003c/h3\u003e\n\u003cp\u003eTo derive functional skin profiles, we applied k-means clustering to 13 standardized indicators. The models for k\u0026thinsp;=\u0026thinsp;2\u0026ndash;6 were fitted with k-means\u0026thinsp;+\u0026thinsp;+\u0026thinsp;initialization, n_init\u0026thinsp;=\u0026thinsp;100, and a fixed random seed. The average silhouette width served as the primary model selection criterion; the Calinski\u0026ndash;Harabasz and Davies\u0026ndash;Bouldin indices and cluster sizes were summarized as secondary diagnostics. The selected number of clusters and diagnostic values are reported in the Results and Supplementary material. The resulting profiles (Profiles 0\u0026ndash;2) served as exposure in the causal analyses.\u003c/p\u003e\n\u003ch3\u003eOutcomes\u003c/h3\u003e\n\u003cp\u003eWe analyzed two outcome families:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eBinary outcomes (expert visual grading): wrinkle, turnover, transparency, seborrheic dermatitis, rosacea, redness, and acne (collapsed to binary per prespecified rules).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eContinuous outcomes (image-derived): Cheek and forehead domains spanning wrinkles, brown spots, red spots, pores, and textures, plus skin color parameters L*, a*, and b*. To report standardized coefficients, each outcome was standardized using the weighted mean and variance of the analytic sample.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e Detailed outcome definitions, image processing, and quality control are provided in the Supplementary Methods.\u003c/p\u003e\n\u003ch3\u003ePropensity score and stabilized IPTW\u003c/h3\u003e\n\u003cp\u003eTo adjust for prespecified confounding, we estimated multinomial propensity scores \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{e}_{a}\\left(X\\right)=\\text{Pr}\\left(A=a\\:\\right|X)\\)\u003c/span\u003e\u003c/span\u003e for the three-level exposure \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:A\\in\\:\\{\\text{0,1},2\\}\\)\u003c/span\u003e\u003c/span\u003e using multinomial logistic regression with age, sex (female=1), skincare habit (yes=1), and current smoking (yes=1) as covariates. The stabilized inverse probability of treatment weights (IPTW) was computed as\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{w}_{i}=\\frac{\\text{P}\\text{r}(A={A}_{i})}{{e}_{{A}_{i}}\\left({X}_{i}\\right)}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ePrimary handling of extreme weights (Option A): We trimmed observations with stabilized IPTW outside the global 1st/99th percentiles and applied this uniformly across outcomes (primary analysis). The realized pooled thresholds are 0.4368 and 2.6360 respectively. Covariate balance was evaluated using absolute standardized mean differences (|SMD|). For a three-level exposure, pairwise |SMD| was summarized as the maximum |SMD| per covariate. Weight distributions (histogram/KDE) and the weighted effective sample size (ESS) after trimming were summarized to characterize the weighting design.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eOutcome models and estimands\u003c/h2\u003e\u003cp\u003eOur estimate was the marginal (population average) effect in the region of common support. We fit weighted marginal structural models (MSMs) with outcome profile indicators (profile 0 as a reference).\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eBinary outcomes: weighted logistic MSMs; effects are reported as odds ratios (OR) with 95% confidence intervals (CI) for Profile 1 vs. 0 and Profile 2 vs. 0.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eContinuous outcomes: weighted linear MSMs; effects are reported as standardized coefficients (β) with 95% confidence intervals CI. Standardization used the weighted outcome s.d. as the denominator.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eMultiple testing correction was applied to functional profiling (Kruskal\u0026ndash;Wallis test with BH-FDR). For downstream outcomes, we prioritized estimation across correlated endpoints, reporting effect sizes, and 95% CIs, rather than significance screening; no across-the-board multiplicity adjustment was applied. Domainwise adjusted summaries can be provided upon request.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eSensitivity analysis\u003c/h3\u003e\n\u003cp\u003eWe applied 1st/99th-percentile winsorization to the stabilized IPTW at the same thresholds as the primary trimming (0.4368/2.6360), without exclusions, and re-fitted all models to assess robustness to the handling of extreme weights. Agreement between the primary trimming and winsorization estimates was quantified using Deming error-in-variables regression (λ\u0026thinsp;=\u0026thinsp;1; accounts for error on both axes) and Lin\u0026rsquo;s concordance correlation coefficient (CCC) together with mean/median absolute errors (MAE/MedAE). For binary outcomes, we compared log odds ratios; for continuous outcomes, we compared the standardized β. We also summarized the proportion of pairwise differences within a region of practical equivalence (ROPE) (\u0026plusmn;\u0026thinsp;0.10 for log-OR; \u0026plusmn;0.05 for β). High CCC and Deming slopes near 1 were pre-specified as evidence that weight handling does not materially distort direction or magnitude.\u003c/p\u003e\n\u003ch3\u003eMissing data\u003c/h3\u003e\n\u003cp\u003eClustering and propensity score estimation used complete cases for the 13 indicators and confounders (N\u0026thinsp;=\u0026thinsp;165). Outcome-specific analyses used non-missing observations after weighting and trimming; denominators therefore vary by outcome and are reported in Figures/tables (see Supplementary Table \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003e for counts and ESS).\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eSoftware\u003c/h2\u003e\u003cp\u003eAnalyses were performed in Python (3.11) using pandas, scikit-learn (clustering), statsmodels (weighted MSMs, robust SE), and matplotlib for visualization. Random seeds were fixed to enhance the reproducibility. Code and exact package versions are listed in the code availability statement.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eEthics\u003c/h2\u003e\u003cp\u003e The study was approved by Hokkaido Information University (approval number: 2023-14). All procedures adhered to the Declaration of Helsinki and written informed consent was obtained from all participants.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThis section describes the three functional skin profiles derived from unsupervised clustering. It then confirmed the validity of the causal inference methodology before presenting the marginal effects of these profiles on a range of binary and continuous clinical phenotypes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.1 Functional profiling and between-group differences\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThirteen skin-function indicators (hydration, transepidermal water loss [TEWL], lipid-related measures, structural markers, and temperature) were z-standardized and clustered using k-means (k = 2\u0026ndash;6; k-means++; n_init = 100; fixed seed). The average silhouette width peaked at k = 3 (0.127) and slightly decreased at k = 4 (0.123), whereas the Calinski\u0026ndash;Harabasz index decreased (31.49\u0026rarr;27.51) and the Davies\u0026ndash;Bouldin index improved (2.204\u0026rarr;1.906) from k = 3 to k = 4 (Supplementary Fig. S1). Therefore, we retained k = 3 for the primary analyses. The cluster size was 70/56/39 (N = 165). The unweighted baseline characteristics (All and Profiles 0\u0026ndash;2) are summarized in Supplementary Table S1. Functional characteristics were visualized using radar plots (Fig. 2a), PCA (Fig. 2b), Cohen\u0026rsquo;s d heat map (Fig. 2c), and indicator ranking by max |d| (Fig. 2d). Formal comparisons used Kruskal\u0026ndash;Wallis with Benjamini\u0026ndash;Hochberg FDR; pairwise effect sizes (d for 0 vs. 1, 0 vs. 2, and 1 vs. 2) and max |d| ranks are provided in Supplementary Table S2. Because we separated function (putative causes) from phenotype (downstream outcomes), these differences are presented as descriptive contexts; causal estimates are reported as marginal effects below. Based on these functional signatures, we hereafter refer to Profile 2 as the \u0026quot;hyperseborrheic\u0026quot; profile, Profile 1 as the \u0026quot; barrier-impaired \u0026quot; profile, and Profile 0 as the \u0026quot;balanced\u0026quot; reference profile.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Covariate balance and weight diagnostics (stabilized IPTW)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMultinomial propensity scores were estimated for the three-level exposure (Profiles 0\u0026ndash;2) using prespecified covariates (age, sex, skincare habit, and current smoking). Stabilized IPTW achieved excellent covariate balance; in the Love plot (Fig. 3a), the maximum pairwise |SMD| per covariate fell within the 0.1 threshold post-weighting. The weight distribution (Fig. 3b) was centered near 1.0; primary analyses applied global 1st/99th-percentile trimming (Option A) with realized pooled thresholds of 0.4368 and 2.6360. Detailed weight summaries are provided in Supplementary Tables S3a (trim) and S3b (winsorization sensitivity). The outcome-specific denominators and weighted effective sample sizes (ESS) after trimming are summarized in Supplementary Table S4, together with weight statistics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Binary phenotypes: clinical interpretation (stabilized IPTW with Option A)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFig. 4 shows the marginal odds ratios from weighted logistic MSMs after stabilized IPTW with primary trimming (Option A). The clinical findings were consistent.\u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eProfile 2: Age-related changes come to the fore, with consistently elevated wrinkle risk. Inflammatory/seborrheic features, such as redness, acne, and seborrheic dermatitis, also tended to be higher.\u003c/li\u003e\n \u003cli\u003eProfile 1: Loss of transparency and slowed turnover dominate, and seborrheic dermatitis is also higher, consistent with dryness, low-grade inflammation, and barrier compromise.\u003c/li\u003e\n \u003cli\u003eProfile 0: binary endpoints for inflammation and ageing remain comparatively low.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eTrends were visualized via a forest plot (Fig. 4a), log(OR) heat map (Fig. 4b), and profile-wise rankings (Fig. 4c\u0026ndash;d). Full estimates are provided in the supplementary Table5.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Continuous phenotypes: focused domains of texture and tone (stabilized IPTW with Option A)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFig. 5 reports standardized coefficients (\u0026beta;) from weighted linear MSMs for five representative domains\u0026mdash;wrinkle, pigmentation, pore, texture, and red spot; the remaining continuous outcomes are summarized in Supplementary Table S6. The clinical read-out is:\u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eProfile 2: Worsening of pore counts and texture roughness was most pronounced, followed by deterioration of wrinkles and pigmentation. These texture-related changes tend to move in the same direction across both the cheek and forehead.\u003c/li\u003e\n \u003cli\u003eProfile 1: Pigmentation (spots) tends to increase, with red spots (mottled erythema) and texture deterioration standing out; the primary signal is a dulling of overall skin tone.\u003c/li\u003e\n \u003cli\u003eProfile 0: these domains remain comparatively favorable.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe direction and relative magnitude are conveyed by forest plots (Fig. 5a, a\u0026prime;) and heat maps (Fig. 5b, b\u0026prime;), and the rankings (Fig. 5c\u0026ndash;d, c\u0026prime;\u0026ndash;d\u0026prime;) indicate which measures most strongly characterize each profile.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Sensitivity analysis: robustness to extreme weights (Option A)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEstimates from primary trimming closely tracked those from winsorization across both binary (log-OR) and continuous (standardized \u0026beta;) outcomes (Fig. 6). Deming slopes were close to unity with small intercepts, Lin\u0026rsquo;s CCC indicated excellent concordance, and absolute errors were modest. The majority of pairwise differences fell within the ROPE, supporting the insensitivity of the conclusions to the specific weight-handling rule. Detailed agreement metrics and side-by-side estimates are provided in Supplementary Materials.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study introduces Dermainformatics, a function-first, causality-oriented framework that separates skin function from phenotype, and quantifies the population-average (marginal) effects of functional profiles on clinically salient outcomes. By deriving profiles from objective biophysical and biochemical indicators\u0026mdash;rather than from visible attributes\u0026mdash; we ground skin typing in measurable physiology and then connected that physiology to phenotypic risk using stabilized IPTW.\u003c/p\u003e\u003cp\u003eA central contribution is the directional separation of predictors (functions) from outcomes (phenotypes). Profiles were defined solely from 13 functional indicators (hydration, barrier loss via TEWL, lipid composition and sebum, structural markers, and temperature) and only thereafter linked to clinical endpoints. This design respects causal temporality, reduces outcome-dependent circularity in classification, and enables the estimation of meaningful interventional marginal effects under standard assumptions (consistency, positivity, and no unmeasured confounding).\u003c/p\u003e\u003cp\u003eThe resulting profiles were mapped to distinct clinical images. Profile 2 (hyperseborrheic) shows the clearest aging-linked signature, with wrinkle risk most consistently elevated and texture roughness and pore counts most pronounced [21], followed by worsening pigmentation, which tends to move in the same direction across the cheek and forehead. In parallel, inflammation/seborrheic features (redness, acne, and seborrheic dermatitis) also tended to be higher, consistent with sebaceous activity, microbial metabolites (e.g., porphyrins) [22, 23], and low-grade inflammation that can accelerate extrinsic aging pathways and collagen degradation [24]. Profile 1 (barrier-impaired) centers on loss of transparency and slowed turnover, with pigmentation (spots), red spots (mottled erythema), and texture deterioration, and the dominant clinical signal is dulling of the overall tone[25, 26]. These patterns align with the reduced hydration/barrier integrity and microinflammation. Profile 0 remained comparatively favorable across domains and served as a functional reference.\u003c/p\u003e\u003cp\u003eMethodologically, Dermainformatics extends skin typing beyond description in two ways. First, weighting-based confounding control (stabilized IPTW) targets marginal effects that are directly actionable for prevention and population risk management; diagnostics confirmed a good balance (|SMD|\u0026lt;0.1), suitable weight distributions, and robustness to the treatment of extreme weights. In the primary analysis, we applied global 1st/99th-percentile trimming (Option A) to restrict inference to the region of covariate overlap; winsorization at the same percentiles (no exclusions) served as a sensitivity analysis, and the conclusions were concordant [15, 27, 28]. Second, the framework pairs biophysical metrics with specimen-derived biochemical markers (e.g., DSG1, fatty acid composition), capturing both the quantity and quality of barrier/sebaceous function, and improving the physiological interpretability of clusters[19, 29].\u003c/p\u003e\u003cp\u003eThese findings have practical implications for personalized skincare and trial design. For Profile 2, programs prioritizing lipid regulation, antioxidant/anti-inflammatory strategies, and texture/pore management may mitigate both inflammatory burden[22] and aging-linked texture/wrinkle trajectories [30]. For Profile 1, barrier restoration and turnover support (e.g., humectants, barrier lipids, gentle keratolytics, and anti-inflammatory care) may address the core \u0026ldquo;dullness\u0026rdquo; phenotype while reducing erythema heterogeneity[25]. Because the effects are marginal and estimated within overlap, these recommendations translate naturally to population-level guidance and enrichment strategies in interventional studies.\u003c/p\u003e\u003cp\u003eThis study has some limitations. The design is observational and cross-sectional; causal interpretation relies on assumptions and measured confounders (age, sex, skincare habit, and smoking)[31, 32]. Generalizability is bounded by the study population (healthy Japanese participants aged 18\u0026ndash;39 years) and by site-specific measurements and device choices. Unsupervised clustering entails modeling decisions (k selection, standardization) that we probed with diagnostics, but remain open to alternative formulations. Finally, while profiles are physiologically interpretable, they are not diagnostic labels; they summarize functional states that can be modified or co-occur.\u003c/p\u003e\u003cp\u003eFuture work should test stability across populations, seasons, and devices, assess longitudinal transitions between functional states, and embed interventional trials that target modifiable functions (barrier, lipids, and micro-inflammation) to verify risk reduction. Integrating multi-omics, microbiome, and digital lifestyle data will likely refine profiles and enhance prediction, while preserving the causal separation of function from phenotype.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eDermainformatics reframes skin typing as a mechanism-aligned functional profiling coupled with marginal causal effects on phenotypes. By separating the function from the outcome, weighting confounders with stabilized IPTW, and validating with diagnostics and sensitivity analyses, the framework yields interpretable, actionable risk estimates. Profile 2 (hyperseborrheic) was most associated with aging-related deterioration, with pronounced texture/pore changes and consistent wrinkle elevation, whereas Profile 1(barrier-impaired) was dominated by tone dullness (transparency loss, slowed turnover) alongside pigmentation and red-spot heterogeneity. These insights support profile-specific prevention and provide a scalable template for precision dermatology that can be extended using longitudinal and multimodal data.\u003c/p\u003e\n\u003cp\u003eDermainformatics reframes skin typing as a generalizable analytical framework in computational dermatology. By separating biological functions from phenotypes and coupling them with stabilized IPTW, it yields population-level, interpretable risk estimates that support risk stratification and personalized recommendation design. Rather than emphasizing immediate clinical deployment, our contribution is a reproducible, physiology-grounded basis for quantitative skin typing that can scale across cohorts and devices. Future work will extend to longitudinal validation, external cohorts, and multimodal integration (e.g., microbiome and lifestyle sensors) while preserving the causal separation of function from phenotype.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe de-identified data that support the findings of this study are available from the corresponding author upon reasonable request for peer review, within the terms of participant consent and applicable data-use agreements. Upon acceptance, the final dataset will be deposited in an open repository and a persistent DOI will be provided in the published version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eReproducible analysis code (Python; exact package versions listed in Methods) will be provided to editors and reviewers via a private repository during peer review. Upon acceptance, the code and environment files will be released under an open-source license and archived with a persistent DOI. Code and minimal reproducible materials will be made publicly available upon acceptance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Cross-ministerial Strategic Innovation Promotion Program (SIP), \u0026ldquo;Building a sustainable food chain that provides abundant and nutritious food\u0026rdquo; (funding agency: Bio-oriented Technology Research Advancement Institution).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was primarily supported by KOS\u0026Eacute; Corporation. The Cross-ministerial Strategic Innovation Promotion Program (SIP), \u0026ldquo;Building a sustainable food chain that provides abundant and nutritious food(tentative)\u0026rdquo; (funding agency: Bio-oriented Technology Research Advancement Institution) provided partial support limited to data acquisition (instrument procurement and field logistics), as this research serves as a foundational step in a broader initiative to explore the link between nutritional status and skin health. Funders had no role in the design, analysis, interpretation, or publication decision.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eN.R. conceived the study, designed the methodology, performed the data analysis, and wrote the original manuscript. N.R. is the lead and corresponding author. K.K. and Y.N. contributed to the study design, performed the on-site data acquisition, and curated the data. A.T. contributed to the study of design and data management. C.Y., K.S., and S.S. conducted the analysis of the samples collected. All authors have read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eEilers, S., et al., \u003cem\u003eAccuracy of Self-report in Assessing Fitzpatrick Skin Phototypes I Through VI.\u003c/em\u003e JAMA Dermatology, 2013. \u003cstrong\u003e149\u003c/strong\u003e(11): p. 1289-1294.\u003c/li\u003e\n\u003cli\u003eSommers, M.S., et al., \u003cem\u003eAre the Fitzpatrick Skin Phototypes Valid for Cancer Risk Assessment in a Racially and Ethnically Diverse Sample of Women?\u003c/em\u003e Ethn Dis, 2019. \u003cstrong\u003e29\u003c/strong\u003e(3): p. 505-512.\u003c/li\u003e\n\u003cli\u003eBaumann, L., \u003cem\u003eUnderstanding and treating various skin types: the Baumann Skin Type Indicator.\u003c/em\u003e Dermatol Clin, 2008. \u003cstrong\u003e26\u003c/strong\u003e(3): p. 359-73, vi.\u003c/li\u003e\n\u003cli\u003eLee, Y.B., et al., \u003cem\u003eWhich Skin Type Is Prevalent in Korean Post-Adolescent Acne Patients?: A Pilot Study Using the Baumann Skin Type Indicator.\u003c/em\u003e Ann Dermatol, 2017. \u003cstrong\u003e29\u003c/strong\u003e(6): p. 817-819.\u003c/li\u003e\n\u003cli\u003eBerardesca, E., \u003cem\u003eEEMCO guidance for the assessment of stratum corneum hydration: electrical methods.\u003c/em\u003e Skin Res Technol, 1997. \u003cstrong\u003e3\u003c/strong\u003e(2): p. 126-32.\u003c/li\u003e\n\u003cli\u003eBerardesca, E., et al., \u003cem\u003eThe revised EEMCO guidance for the in vivo measurement of water in the skin.\u003c/em\u003e Skin Res Technol, 2018. \u003cstrong\u003e24\u003c/strong\u003e(3): p. 351-358.\u003c/li\u003e\n\u003cli\u003eRogiers, V., \u003cem\u003eEEMCO guidance for the assessment of transepidermal water loss in cosmetic sciences.\u003c/em\u003e Skin Pharmacol Appl Skin Physiol, 2001. \u003cstrong\u003e14\u003c/strong\u003e(2): p. 117-28.\u003c/li\u003e\n\u003cli\u003eSherrill, J.D., et al., \u003cem\u003eDesmoglein-1 regulates esophageal epithelial barrier function and immune responses in eosinophilic esophagitis.\u003c/em\u003e Mucosal Immunol, 2014. \u003cstrong\u003e7\u003c/strong\u003e(3): p. 718-29.\u003c/li\u003e\n\u003cli\u003eDel Rosso, J.Q. and L. Kircik, \u003cem\u003eThe primary role of sebum in the pathophysiology of acne vulgaris and its therapeutic relevance in acne management.\u003c/em\u003e Journal of Dermatological Treatment, 2024. \u003cstrong\u003e35\u003c/strong\u003e(1): p. 2296855.\u003c/li\u003e\n\u003cli\u003eCao, K., et al., \u003cem\u003eFatty Acid Profiling in Facial Sebum and Erythrocytes From Adult Patients With Moderate Acne.\u003c/em\u003e Frontiers in Physiology, 2022. \u003cstrong\u003eVolume 13 - 2022\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eCzajkowska, J., et al., \u003cem\u003eHigh-frequency ultrasound in anti-aging skin therapy monitoring.\u003c/em\u003e Scientific Reports, 2023. \u003cstrong\u003e13\u003c/strong\u003e(1): p. 17799.\u003c/li\u003e\n\u003cli\u003eMeng, Y., et al., \u003cem\u003eApplication of high-frequency ultrasound to assess facial skin thickness in association with gender, age, and BMI in healthy adults.\u003c/em\u003e BMC Med Imaging, 2022. \u003cstrong\u003e22\u003c/strong\u003e(1): p. 113.\u003c/li\u003e\n\u003cli\u003eMacRae, B.A., et al., \u003cem\u003eSkin Temperature Measurement Using Contact Thermometry: A Systematic Review of Setup Variables and Their Effects on Measured Values.\u003c/em\u003e Frontiers in Physiology, 2018. \u003cstrong\u003eVolume 9 - 2018\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eRobins, J.M., M.\u0026Aacute;. Hern\u0026aacute;n, and B. Brumback, \u003cem\u003eMarginal Structural Models and Causal Inference in Epidemiology.\u003c/em\u003e Epidemiology, 2000. \u003cstrong\u003e11\u003c/strong\u003e(5).\u003c/li\u003e\n\u003cli\u003eHern\u0026aacute;n, M.A., B. Brumback, and J.M. Robins, \u003cem\u003eMarginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men.\u003c/em\u003e Epidemiology, 2000. \u003cstrong\u003e11\u003c/strong\u003e(5): p. 561-70.\u003c/li\u003e\n\u003cli\u003eGreenland, S., \u003cem\u003eNoncollapsibility, confounding, and sparse-data bias. Part 1: The oddities of odds.\u003c/em\u003e J Clin Epidemiol, 2021. \u003cstrong\u003e138\u003c/strong\u003e: p. 178-181.\u003c/li\u003e\n\u003cli\u003eGreenland, S., \u003cem\u003eNoncollapsibility, confounding, and sparse-data bias. Part 2: What should researchers make of persistent controversies about the odds ratio?\u003c/em\u003e Journal of Clinical Epidemiology, 2021. \u003cstrong\u003e139\u003c/strong\u003e: p. 264-268.\u003c/li\u003e\n\u003cli\u003eMcCaffrey, D.F., et al., \u003cem\u003eA tutorial on propensity score estimation for multiple treatments using generalized boosted models.\u003c/em\u003e Stat Med, 2013. \u003cstrong\u003e32\u003c/strong\u003e(19): p. 3388-414.\u003c/li\u003e\n\u003cli\u003eNaoe, Y., et al., \u003cem\u003eBidimensional analysis of desmoglein 1 distribution on the outermost corneocytes provides the structural and functional information of the stratum corneum.\u003c/em\u003e J Dermatol Sci, 2010. \u003cstrong\u003e57\u003c/strong\u003e(3): p. 192-8.\u003c/li\u003e\n\u003cli\u003eCamera, E., et al., \u003cem\u003eUse of lipidomics to investigate sebum dysfunction in juvenile acne [S].\u003c/em\u003e Journal of Lipid Research, 2016. \u003cstrong\u003e57\u003c/strong\u003e(6): p. 1051-1058.\u003c/li\u003e\n\u003cli\u003eRoh, M., et al., \u003cem\u003eSebum output as a factor contributing to the size of facial pores.\u003c/em\u003e Br J Dermatol, 2006. \u003cstrong\u003e155\u003c/strong\u003e(5): p. 890-4.\u003c/li\u003e\n\u003cli\u003eSpittaels, K.J., et al., \u003cem\u003ePorphyrins produced by acneic Cutibacterium acnes strains activate the inflammasome by inducing K(+) leakage.\u003c/em\u003e iScience, 2021. \u003cstrong\u003e24\u003c/strong\u003e(6): p. 102575.\u003c/li\u003e\n\u003cli\u003eMayslich, C., P.A. Grange, and N. Dupin, \u003cem\u003eCutibacterium acnes as an Opportunistic Pathogen: An Update of Its Virulence-Associated Factors.\u003c/em\u003e Microorganisms, 2021. \u003cstrong\u003e9\u003c/strong\u003e(2).\u003c/li\u003e\n\u003cli\u003eShimizu, N., et al., \u003cem\u003eOxidation of squalene by singlet oxygen and free radicals results in different compositions of squalene monohydroperoxide isomers.\u003c/em\u003e Scientific Reports, 2018. \u003cstrong\u003e8\u003c/strong\u003e(1): p. 9116.\u003c/li\u003e\n\u003cli\u003eGreen, M., et al., \u003cem\u003eTransepidermal water loss (TEWL): Environment and pollution-A systematic review.\u003c/em\u003e Skin Health Dis, 2022. \u003cstrong\u003e2\u003c/strong\u003e(2): p. e104.\u003c/li\u003e\n\u003cli\u003eGalizia, G., A. Belloni Fortina, and A. Semenzato, \u003cem\u003eSeborrheic Dermatitis: From Microbiome and Skin Barrier Involvement to Emerging Approaches in Dermocosmetic Treatment.\u003c/em\u003e Cosmetics, 2024. \u003cstrong\u003e11\u003c/strong\u003e(6): p. 208.\u003c/li\u003e\n\u003cli\u003eRobins, J.M., M.A. Hern\u0026aacute;n, and B. Brumback, \u003cem\u003eMarginal structural models and causal inference in epidemiology.\u003c/em\u003e Epidemiology, 2000. \u003cstrong\u003e11\u003c/strong\u003e(5): p. 550-60.\u003c/li\u003e\n\u003cli\u003eCole, S.R. and M.A. Hern\u0026aacute;n, \u003cem\u003eConstructing Inverse Probability Weights for Marginal Structural Models.\u003c/em\u003e American Journal of Epidemiology, 2008. \u003cstrong\u003e168\u003c/strong\u003e(6): p. 656-664.\u003c/li\u003e\n\u003cli\u003eCohen-Barak, E., et al., \u003cem\u003eThe Role of Desmoglein 1 in Gap Junction Turnover Revealed through the Study of SAM Syndrome.\u003c/em\u003e J Invest Dermatol, 2020. \u003cstrong\u003e140\u003c/strong\u003e(3): p. 556-567.e9.\u003c/li\u003e\n\u003cli\u003eNakamura, R., et al., \u003cem\u003ePrediction of future wrinkles for middle-aged women: A 7-year longitudinal study on the progression of wrinkles in Japanese women.\u003c/em\u003e Skin Research and Technology, 2021. \u003cstrong\u003e27\u003c/strong\u003e(5): p. 854-862.\u003c/li\u003e\n\u003cli\u003eAustin, P.C. and E.A. Stuart, \u003cem\u003eMoving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.\u003c/em\u003e Stat Med, 2015. \u003cstrong\u003e34\u003c/strong\u003e(28): p. 3661-79.\u003c/li\u003e\n\u003cli\u003eAustin, P.C., \u003cem\u003eBalance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples.\u003c/em\u003e Stat Med, 2009. \u003cstrong\u003e28\u003c/strong\u003e(25): p. 3083-107.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"computational dermatology, skin informatics, causal inference, inverse probability weighting, risk stratification, personalized skincare","lastPublishedDoi":"10.21203/rs.3.rs-8013431/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8013431/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eConventional skin type classifications rely on subjective, phenotype-based descriptors (e.g., dryness, oiliness, sensitivity), limiting mechanistic insight and predictive value. We introduce Dermainformatics, novel data-driven framework designed to separate biological skin function from its clinical phenotype and embed causal inference. Thirteen biophysical/biochemical indicators (e.g., stratum corneum hydration, transepidermal water loss (TEWL), desmoglein-1 (DSG1), sebum and fatty-acid composition, collagen content, skin thickness, skin temperature) were clustered using k-means with silhouette-based model selection, yielding three functional profiles (N\u0026thinsp;=\u0026thinsp;165). We then estimated marginal (population-average) effects of profiles on expert-graded clinical and appearance-related binary outcomes and image-derived continuous measures using stabilized inverse probability of treatment weighting (IPTW) with prespecified confounders (age, sex, skincare habit, current smoking). Primary analyses applied global 1st/99th-percentile trimming of stabilized IPTW (Option A); robustness was evaluated with covariate-balance diagnostics (|SMD|\u0026lt;0.1), weight distributions and effective sample size, and a winsorization sensitivity at the same percentiles. Profiles exhibited distinct functional signatures with coherent, population-average associations for wrinkles, redness, transparency, acne, pores, and porphyrin features (reported as odds ratios and standardized β). By decoupling function from phenotype and quantifying marginal causal effects within the region of covariate overlap, Dermainformatics yields interpretable, mechanism-aligned stratification suitable for personalized skincare and population-level risk prediction, and is readily extensible to multi-omics and lifestyle data for precision prevention. This framework provides a reproducible, physiology-grounded basis for quantitative skin typing and population-level risk estimation. As a generalizable contribution to computational dermatology, Dermainformatics facilitates preventive and personalized skincare by coupling function-based profiling with causal inference.\u003c/p\u003e","manuscriptTitle":"Dermainformatics: A Data-Driven Framework for Function-Based Skin Typing and Causal Estimation of Skin Health Outcomes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-26 06:27:29","doi":"10.21203/rs.3.rs-8013431/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"94950032-4703-4c10-af4b-e7db852e2883","owner":[],"postedDate":"November 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":58452484,"name":"Health sciences/Biomarkers"},{"id":58452485,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":58452486,"name":"Health sciences/Diseases"},{"id":58452487,"name":"Health sciences/Health care"},{"id":58452488,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-12-16T12:09:17+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-26 06:27:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8013431","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8013431","identity":"rs-8013431","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}