ETD

• Biomimetic sensing
• Cheminformatics
• Data science
• Machine learning
• Molecular docking
• Predictive screening
• Sensor array

This thesis presents an interdisciplinary study at the intersection of data science and biomimetic sensing, conducted during an internship at the Biomimetic Functional Materials and Sensing Laboratory. The work is structured around three interconnected projects. First, machine learning models—including Linear Discriminant Analysis, Logistic Regression, and Decision Trees—were applied to classify six structurally diverse bitter molecules using response data from a four-membrane sensor array functionalised with the TAS2R38 receptor. The models achieved perfect classification accuracy, validating the discriminative power of the array and revealing a synergistic mechanism between specific and non-specific sensing channels. Second, a molecular docking workflow was developed to investigate the recognition mechanisms underlying this discrimination. By integrating spatial filtering with literature-guided validation, key residues involved in ligand binding were identified, offering a foundation for future peptide-based sensor design. Third, a dual-layer computational screening framework was developed to
predict Drosophila olfactory receptor activation by herbivore-induced
plant volatiles (HIPVs). Morgan fingerprints and Logistic Regression
classifiers trained on the DoOR olfactory response database were used
to benchmark per-receptor structure--activity signal across a
full molecular library and to evaluate ecological coherence within
green leaf volatile and terpene subclasses. Or13a, Or42a, and Or35a
were identified as statistically reliable candidates for a
GLV-sensitive sensor channel, while Or69a emerged as the strongest
candidate for broad-spectrum HIPV coverage. Together, these projects demonstrate how data science methods---ranging
from classification and interpretability analysis to cheminformatics
pipeline design and structure--activity modelling---can be integrated
into materials science workflows to enhance sensor design, interpretability, and biological relevance.