Singapore's NUS Pioneers AI-Driven Discovery in Sleep-Promoting Aromatics

In a landmark advancement from the National University of Singapore (NUS), researchers have harnessed artificial intelligence (AI) to unlock the sleep-inducing potential hidden within nearly 1,000 aromatic plants. This study, published in the prestigious Digital Discovery journal, systematically maps the sleep-promoting effects of 2,391 volatile organic compounds (VOCs)—naturally occurring scent molecules—extracted from 991 diverse aromatic species. Led by Assistant Professor Dr. Dachuan Zhang from NUS's Department of Food Science and Technology, the work represents a fusion of food informatics, machine learning, and pharmacology, offering a blueprint for natural alternatives to synthetic sleep aids.

Sleep disorders plague up to one-third of the global population, with pharmaceutical interventions like benzodiazepines often leading to dependency, cognitive impairment, and other side effects. Traditional aromatherapy, relying on plants like lavender, has anecdotal support but lacks rigorous scientific mapping. The NUS team's AI approach bridges this gap, predicting sleep efficacy with unprecedented accuracy and validating predictions through in vivo experiments.

Unpacking the Methodology: AI Meets Plant Chemistry

The core innovation lies in an ensemble machine learning framework that analyzes molecular structures of VOCs to forecast their sleep-promoting activity. Data was curated from extensive databases of plant volatiles, encompassing structural descriptors such as MACCS keys (a binary fingerprint system representing molecular substructures), RDKit fingerprints (topological and physicochemical features), and ECFP4 (extended-connectivity fingerprints capturing circular neighborhoods around atoms).

Multiple models were trained and stacked: Random Forest (RF), XGBoost (gradient boosting), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and graph neural networks like AttentiveFP, MPNN, CHEM-BERTa, and KANO. The stacking ensemble—combining RF-MACCS, RF-RDKit, XGBoost-MACCS, and SVM-MACCS—achieved a stellar area under the curve (AUC) of 0.994 and accuracy of 0.961 on five-fold cross-validation, far surpassing individual models.

This rigorous process involved positive samples (known sleep-promoting VOCs) and negatives, ensuring balanced training. Five top-predicted candidates were selected for mouse model testing, measuring sleep via electroencephalogram (EEG) for non-rapid eye movement (NREM) sleep duration and gamma-aminobutyric acid (GABA) receptor expression—key to sedative effects.

Key Discoveries: Standout Volatiles and Their Plant Sources

The AI models pinpointed high-potential VOCs with prediction scores nearing 1.0. Leading the pack is carvacrol (score 1.0), found in plants like oregano (Origanum vulgare), holy basil (Ocimum sanctum), and thyme species (Thymus amurensis, Thymus mongolicus). Next, safranal (1.0) from lotus (Nelumbo nucifera) and purslane (Portulaca oleracea); methyl eugenol (1.0) abundant in laurel family plants like cinnamon (Cinnamomum spp.) and nutmeg (Myristica fragrans); linalool (1.0), the star in lavender (Lavandula angustifolia), perilla (Perilla frutescens), basil (Ocimum basilicum), and rosemary (Rosmarinus officinalis); and vanillin (0.96) from vanilla (Vanilla planifolia).

• Carvacrol: Antifungal and anxiolytic, modulates GABA receptors.
• Linalool: Widely studied for calming effects, enhances NREM sleep.
• Safranal: Antioxidant from saffron relative, promotes relaxation.

These compounds exemplify how AI sifts through chemical complexity to highlight therapeutically promising molecules.

Validation Success: From Prediction to Real-World Proof

Computational hits were rigorously tested in vivo. Of five prioritized VOCs, four (80% hit rate) significantly boosted NREM sleep duration in mice, as confirmed by EEG, and upregulated GABA_A receptor subunits—mirroring pharmaceutical sedatives without toxicity. This validation underscores the model's reliability, bridging silico predictions with biological outcomes.

Such empirical confirmation is rare in natural product screening, positioning this workflow as a gold standard for future discoveries.

Photo by Brett Jordan on Unsplash

Plant Families and Species Poised for Spotlight

Phylogenetic analyses revealed enriched families: Lamiaceae (mint family, e.g., lavender, basil, perilla) dominated with diverse high-scoring VOCs; Lauraceae (laurels, cinnamons); and Asteraceae (daisies, artichokes). Gymnosperms like Pinaceae (pines) and Cupressaceae (cypresses) also showed promise.

Promising species include Lavandula angustifolia (lavender, long-known but now quantified), Perilla frutescens (shiso, common in Asian cuisine), and Ocimum basilicum (basil). These could inspire new essential oil blends or fortified foods.

AI's Transformative Power in Natural Product Research

Dr. Zhang's FoodAI group at NUS exemplifies Singapore's prowess in food informatics. Traditional screening tests thousands of compounds manually; AI reduces this to dozens, slashing time and cost. The stacking ensemble's near-perfect metrics highlight tailored model integration for chemical prediction.Learn more about NUS Food Science research

This aligns with Singapore's National AI Strategy 2.0, investing in AI for biotech and agritech.

Health and Industry Implications: Beyond the Lab

Findings pave the way for sleep-enhancing products: essential oils, herbal teas, or diffusers targeting carvacrol-rich oregano or linalool-heavy lavender. With global sleep aid market at $80B+, natural options could capture wellness niches. GABA modulation suggests pharmaceutical potential without addiction risks.

For consumers, actionable: Incorporate perilla in diets or basil teas. For industry, AI accelerates R&D, from functional foods to cosmetics.

• Potential in higher ed research jobs in food tech.
• Links to academic career advice.

Singapore's Universities Leading AI in Higher Education Research

NUS's role underscores Singapore's higher education edge. Dr. Zhang, IUFoST Young Scientist 2024 awardee, leads FoodAI, decoding food systems via data. Collaborations with Shanghai Institute highlight international ties.

Similar at NTU and A*STAR: AI in drug discovery, sustainable agri. Aspiring researchers: Explore Singapore university jobs.

Photo by Brett Jordan on Unsplash

Challenges, Limitations, and Path Forward

While promising, challenges remain: VOC synergies unmodeled, human trials needed, cultural/regional plant access. Future: Expand to 10,000+ compounds, integrate multi-omics, clinical RCTs.

Reusable workflow could screen for anxiety, cognition aids.

Career Opportunities in Singapore's AI-Biotech Boom

This study spotlights demand for AI-savvy food scientists. NUS, NTU seek postdocs, faculty in informatics. Check higher ed faculty jobs, research assistant roles, rate professors.

Actionable: Pursue winning academic CV tips. Singapore's ecosystem offers scholarships, visas for talents.Read the full NUS study