Transparent Multitask SERS Analysis with Explainable AI
Surface enhanced Raman spectroscopy (SERS) provides a highly sensitive detection of biomolecules; however, interpretation of these high dimensional data is challenging and requires the use of advanced spectra analytical techniques such as machine learning models. These models lack transparency and interpretability, limiting their acceptance in real-world applications. Understanding the rationale behind machine learning predictions would enhance transparency and build confidence among chemical and biomedical researchers. Here, we describe an attention-based multitask convolutional neural network (MCNN) that simultaneously classifies and quantifies three catecholamines (dopamine, epinephrine, norepinephrine) and then apply two explainable AI (XAI) techniques, SHapley Additive exPlanation (SHAP) and Local Interpretable Model Agnostic Explanation (LIME), for interpretation. The classification task achieved 100% accuracy in distinguishing among the catecholamines while the regression task had a mean correlation coefficient of 0.97. LIME and SHAP were successfully applied for a complementary explanation of the model predictions.
Authors
- Bhavya Sharma (ORCID: https://orcid.org/0000-0003-4388-5702)
- Wilson Garuba
Institutions
- University of Tennessee at Knoxville (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-14
- DOI
- https://doi.org/10.5281/zenodo.24221644
- Primary Topic
- Spectroscopy Techniques in Biomedical and Chemical Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00