Machine Learning Data Augmentation for Essential Oil Composition–Activity Modeling
A machine learning and data augmentation approach to model essential oil composition–activity relationships, enabling prediction of biological effects such as telomerase modulation and antibacterial activity. Built on a dataset of 2,767 characterized plant extracts, the method addresses composition variability to support discovery of proprietary essential oil mixtures for nutritional and supplement applications.
Overview
This solution applies machine learning (ML) and deep learning data augmentation (DA) techniques to essential oil (EO) composition–activity modeling. EOs are complex natural mixtures whose chemical profiles vary due to plant source and extraction method, making standardization and predictive modeling difficult. The approach uses DA to dynamically reshape EO composition datasets, addressing class imbalance and enabling robust predictive models for biological activity. Validated previously on antibacterial targets and now being extended to telomerase activity, the method supports discovery of undisclosed properties and the design of proprietary EO mixtures for the nutritional and supplement sectors.
Technical specifications
Key features:
- Quantitative composition–activity relationship (QCAR) models built using ML and deep learning algorithms
- Data augmentation strategy applied to EO chemical composition data to balance datasets and improve classification performance
- Training dataset comprising 2,767 characterized plant extracts with known EO compositions
- Application demonstrated on antibacterial screenings and proposed for telomerase length modulation
- Initial validation set of 70–80 EOs with well-characterized chemical compositions
- Designed to handle high-dimensional, scarce biological data typical of natural product research
Technology readiness level
The data augmentation approach has been validated on antibacterial targets through in vitro screening campaigns. Application to telomerase activity represents a new validation phase, leveraging prior proof-of-feasibility in related biological domains. The method is positioned at an early-to-mid technology readiness stage, with established ML infrastructure and a large reference dataset supporting further development and pilot studies in the nutritional and supplement field.
About Universita degli Studi di Ferrara
The University of Ferrara is a comprehensive public research university in northern Italy with a collaborative, industry‑oriented culture. Industry engagement is anchored by the Tecnopolo di Ferrara, which the university manages as a hub connecting companies with accredited industrial research laboratories and a regional innovation consortium, enabling streamlined contracting and collaboration. Its integrated university hospital system provides a platform for clinical research and translation in partnership with the university. Located in Emilia‑Romagna’s dense manufacturing ecosystem, partners benefit from proximity to suppliers, SMEs, and testing environments across the region. Research is supported by competitive European and national funding. A dedicated technology transfer office guides IP protection, licensing, and startup creation to accelerate commercialization.
delivered to your inbox.