Assessment of scalar and molecular biomarkers in predicting treatment response in onychomycosis using dermal imaging and machine learning

Project Abstract

This study investigates the predictive performance of scalar and molecular biomarkers in forecasting treatment response in onychomycosis through an integrative framework that combines high-resolution dermal imaging, molecular profiling, and machine learning. We recruited a cohort of 240 patients diagnosed with distal and lateral subungual onychomycosis across multiple clinical sites, collecting baseline dermal images, mycological cultures, and nail bed transcriptomic and proteomic data. Treatment regimens consisted of standardized oral antifungals with adjunctive topical therapy, and outcomes were recorded at 12 and 24 weeks using mycological cure, complete clinical response, and 12-month relapse rates. Imaging data underwent preprocessing for color normalization, lesion segmentation, and texture analysis, yielding quantitative features such as pigment density, vascularization indices, and periungual tissue integrity. Molecular data encompassed targeted panels for fungal load indicators, host immune response genes, and keratinocyte signaling pathways, enabling calculation of composite scores for fungal burden, inflammatory milieu, and tissue remodeling activity. Feature selection integrated univariate filtering, recursive feature elimination, and SHAP-based importance metrics to identify robust biomarkers across modalities. Machine learning models including gradient boosting, random forest, support vector machines, and deep learning architectures were trained to predict treatment response categories (cure, partial response, no response) and time-to-response. Model interpretability was enhanced through LIME explanations and attention maps, revealing key interactions between molecular signatures and imaging phenotypes. Our results demonstrate that combining scalar clinical metrics with molecular biomarkers improved predictive accuracy significantly over imaging or molecular data alone. The best-performing model achieved an area under the receiver operating characteristic curve of 0.89 for binary cure vs non-cure prediction at 24 weeks, with a concordance index of 0.82 for time-to-response analysis. Important predictors included high levels of host inflammatory transcripts (e.g., IL1B, IL6) coupled with imaging features of increased nail bed vascularity and heterogeneous pigment deposition, alongside reduced expression of keratinization-related genes and elevated fungal DNA burden indicators. External validation on an independent cohort (n=60) yielded comparable performance, supporting generalizability across diverse populations. Subgroup analyses indicated that biomarker utility varied with clinical subtype, prior treatment history, and demographic factors, suggesting personalized therapeutic guidance. Sensitivity analyses confirmed robustness to missing data and imaging variability, while ablation studies highlighted the synergistic value of multi-omics integration. The study provides a framework for precision dermatology in onychomycosis, enabling early identification of likely responders and non-responders, optimization of treatment duration, and monitoring strategies through noninvasive imaging and targeted molecular profiling. Limitations include potential confounding by concurrent dermatological conditions and the need for cost-effective, scalable workflows for routine clinical deployment. Future work will explore longitudinal multi-omics trajectories, integration with genomic risk scoring, and real-time decision support tools embedded in dermatology clinics.

Project Overview

What This Project Is About

---

A plain-language overview of the topic and what the project investigates.

---

---

The Problem It Addresses

---

What problem or gap this project tackles and why it matters to the field or society.

---

---

Objectives of the Project

---

1. Identify scalar and molecular biomarkers linked with treatment response in onychomycosis.
2. Develop a simple workflow combining imaging data with basic machine learning to predict outcomes.
3. Evaluate whether dermal imaging signals can improve treatment decisions alongside standard care.
4. Explain findings in accessible terms for clinical use and patient care.

---

---

What You Will Do Step by Step

---

1. Review background literature on onychomycosis and biomarkers.
2. Collect or access dermal images and related data from patients or public datasets.
3. Annotate data for treatment outcomes (success, partial response, failure).
4. Extract simple, interpretable features from images and any available molecular data.
5. Train a straightforward predictive model and assess its accuracy.
6. Validate findings with a small independent sample and discuss limitations.

---

---

Expected Outcome

Expect to produce a practical model that uses easy-to-understand biomarkers to forecast how well a treatment will work, along with clear guidelines for clinicians on when to adjust therapy.