–
May 15, 2025
Nathanial Brittin, Pharmsci graduate student (Bugni Research Group), will be defending his PhD research thesis:
Computational Metabolomics for Bioactivity-Driven Classification in Natural Products Drug Discovery and Opioid Detection
Abstract:
The rise of multidrug-resistant (MDR) pathogens and synthetic opioids both pose significant threats to global health, necessitating innovative solutions in drug discovery and forensic analysis. This dissertation presents a multidisciplinary approach utilizing high-resolution mass spectrometry, yeast chemical genomics (YCG), and machine learning (ML) to address these challenges across natural product discovery and clinical toxicology.
To accelerate dereplication of antifungals in drug discovery, we developed a high-throughput screening platform integrating LC-MS/MS-based metabolomics and YCG profiling. Approximately 40,000 bacterial extract fractions from diverse microbiomes were screened for antifungal activity. Hits were prioritized based on chemical-genetic interaction profiles and computational metabolomics. This combined approach enabled functional and structural dereplication, linking antifungal bacterial extracts to known compounds and mechanisms of action, facilitating efficient prioritization of promising lead antifungals.
To further streamline the dereplication process in natural products drug discovery, we trained ML classifiers to categorize bioactive natural products into 21 pharmacophore-defined drug classes using LC/MSMS metabolomics data. These classifiers demonstrated over 93% accuracy in multiclass predictions, significantly outperforming current methods on public experimental data. Application to microbial extracts allowed rapid bioactivity classification without reliance on mass spectral databases, thus expanding the accessible chemical space for antimicrobial discovery.
We then expanded upon the concept of pharmacophore classification and developed ML-based opioid detection models using clinical LC-MS/MS data. Classifiers targeting morphinan, fentanyl, and nitazene opioids were developed using simulated molecular fingerprints and validated with public spectral libraries and anonymized clinical samples. Morphinan and fentanyl classifiers exhibited exceptional performance, improving compound detection by 90–600% and 33–400%, respectively, over conventional spectral database searches. However, the nitazene classifier revealed limitations in clinical settings, highlighting a need for enhanced training data and coverage.
Overall, this research illustrates how integrating metabolomics, machine learning, and functional genomics can transform small molecule discovery and enhance opioid identification, enabling scalable, structure-informed identification of bioactive compounds and synthetic opioids.
