Integrated AI platform for the discovery of plant-derived bioactive compounds

Developing a drug from plant raw material the classical way is long and costly. The route runs from an ethnopharmacological hypothesis, through blind screening of hundreds of extracts, to chance «hits» and years of optimization. Our platform shortens that route. It brings ethnopharmacology, analytical chemistry, chemoinformatics, machine learning and experimental validation together into a single sequential process.

Here, artificial intelligence does not replace the researcher. It supports decision-making and helps narrow the search for promising candidates. The ultimate goal is the shift from «blind phytotherapy» to data-driven phytopharmaceutics.

1. Forming the ethnopharmacological hypothesis

We start by forming a hypothesis for a specific therapeutic direction — metabolic diseases, cardiovascular pathologies, inflammatory and neuroinflammatory processes. Our sources are ethnopharmacological and ethnobotanical literature, modern phytochemical and pharmacological studies, established traditional practices of plant use, and bioinformatic and chemoinformatic approaches. At this stage we weigh therapeutic relevance and the availability of raw material in Ukraine. We also account for seasonal and geographic variability of composition, outline potential targets for the action of bioactive compounds, and look for prior safety data.

2. AI analysis of the scientific literature and expert validation

We analyze the literature with AI agents and ML tools. They work with PubMed, ChEMBL, open phytochemical databases, ethnopharmacological sources and natural-compound databases. The result is a structured profile: known bioactive substances, their activity, possible molecular targets, extraction methods and available toxicological data. An expert always verifies these data, since AI models can produce inaccurate or incomplete interpretations of phytochemical information.

3. Raw material fingerprinting and authentication

The experimental part begins with fingerprinting the raw material by physico-chemical methods — chromatographic (LC-MS, GC), spectral and others, as appropriate. The aim is to confirm the authenticity of the plant raw material and to build a standardized chemical profile. Along the way we detect signals that may correspond to marker components, control batch-to-batch reproducibility, and check the material for impurities or adulteration. We pay particular attention to seasonal and geographic variability of composition, since the concentration of active substances depends substantially on the growing conditions.

4. Extraction optimization (DoE + AI)

We optimize extraction using Design of Experiments (DoE), statistical modeling and AI analysis of the resulting data. We vary solvents, temperature, pH, extraction time and the solid-phase / solvent ratio. AI does not build the protocol automatically. Instead, it analyzes series of experiments and suggests optimal parameters for obtaining physico-chemical profiles with the desired activity.

5. In silico activity prediction and multi-target analysis

For the identified compounds we apply a hybrid approach: ML prediction of biological activity, pharmacophore analysis, ADMET prediction and chemoinformatic expert assessment. We are most interested in multi-target effects, potential synergy, toxicity prediction and how compounds interact within mixtures. Most ML models are trained mainly on individual synthetic molecules. That is why we complement prediction for natural mixtures with expert interpretation and experimental validation.

6. Design and optimization of «hit» compositions

The classical approach works on the «one molecule — one target» principle. Our platform aims instead at rational multi-target compositions. Such mixtures offer several advantages. They act on several pathogenetic mechanisms at once, produce synergistic effects, and reduce the risk of resistance and toxicity. They also allow lower concentrations of individual constituents and fit complex chronic diseases better. The approach is especially promising for metabolic syndrome, type 2 diabetes, chronic inflammation, cardiovascular and neuroinflammatory pathologies, as well as antimicrobial and antibiofilm systems. This is the concept of AI-guided precision phytopharmacology.

7. Experimental in vitro validation

A short list of priority compounds, standardized extracts and «hit» compositions moves on to the experimental stage. Here we run cell-based tests, enzymatic assays, cytotoxicity evaluation, mechanism-of-action analysis and synergy verification of the constituents. This lets us filter out random candidates and focus resources on the most promising compositions.

8. Preclinical translation and application prospects

The platform's main output is a standardized plant-derived «hit» mixture with a controlled composition, characterized by physico-chemical methods. Such a mixture can be patented or transferred to pharmaceutical companies as «know-how» for the next stages of drug development. It can also serve as the basis for a new generation of dietary supplements. Promising areas of application include metabolic and cardiometabolic diseases, neuroinflammatory conditions, microbiome modulation, antioxidant systems, adjuvant therapy of chronic diseases, and antimicrobial and antibiofilm compositions.

Why this approach is promising

The platform saves time and resources at primary screening and moves us from «blind phytotherapy» toward data-driven phytopharmaceutics. Standardized physico-chemical profiles, together with the integration of ethnopharmacology, metabolomics, AI and experimental pharmacology, yield the concept of «controlled mixtures». This is no longer a classical «extract of unknown composition», but a standardized composition with a controlled chemical profile and predicted activity.

The key advantage is the move away from the «one molecule — one target» model toward the more realistic multi-target mode of action. For metabolic, inflammatory, neurodegenerative and cardiovascular diseases this matters a great deal. Such pathologies are almost never linked to a single signaling pathway. In metabolic diseases, oxidative stress, chronic inflammation, insulin resistance, microbiome disturbances, mitochondrial dysfunction and vascular disorders are present at the same time. A single molecule rarely covers this whole spectrum. A rationally designed phytocomposition, by contrast, can act on several pathogenetic nodes at once.

Combining ethnopharmacology, analytical chemistry, metabolomics, machine learning and experimental pharmacology within a single workflow lets us search for bioactive plant compositions in a systematic and reproducible way. This interdisciplinary format opens the way to standardized multi-target compositions with a controlled composition and predicted activity.

Publications

1. Maiula TG (2026) LC-MS chromatographic fingerprinting as a tool for medicinal plant standardization (Hypericum perforatum L. as a case study). ScienceRise: Biological Science. https://doi.org/10.15587/2519-8025.2026.362775
2. Khromov A, Dobrelia N, Konvalyuk I, Bieda O, Andreev I, Mozhylevska L, Yarmoluk S, Soloviev A, Kunakh V (2025) Alkaloid Composition and Antiarrhythmic Activity of the Extracts from Rauvolfia serpentina Tissue Culture. Planta Med 91(12):645-655.
3. Maiula TG, Yarmoliuk SM, Konvaliuk II, Kunakh VA (2025) In silico prediction of neuroprotective properties of natural compounds using Scutellaria baicalensis as an example. Biopolym Cell 41(2):139-149. https://doi.org/10.7124/bc.000B18
4. Maiula TG, Bieda OA, Pylaieva TV, Yarmoluk SM (2025) In silico modeling and prediction of antidiabetic potential of bioactive compounds from Galega officinalis L. non-alkaloid extract. Biopolym Cell 41(4):309-317. https://doi.org/10.7124/bc.000B29
5. Stepanova SI, Strygol SYu, Kotsar OO, Bieda OA, Konvalyuk II, Pylaeva TV, Mozhylevska LP, Yarmoluk SM, Kunakh VA (2025) Chemical composition and effect of Ungernia victoris tissue culture extract on behavioral responses, movement coordination, and pain sensitivity in mice. Pharmacol Drug Toxicol 19(2).
6. Khromov OS, Dobrelia NV, Parshikov OV, Konvalyuk II, Bieda OA, Mozhylevska LP, Yarmoluk SM, Soloviev AI, Kunakh VA (2023) The hypotensive activity of different fractions of the extracts from tissue culture biomass Rauwolfia serpentina Benth. Fiziol Zh 69(4):45-53.
7. Kunakh VA, Konvalyuk II, Mozhylevska LP, Bieda OA, Twardovska MO, Andreev IO, Yarmolyuk SM (2023) Comprehensive study of hormone-independent highly productive strain of Rauvolfia serpentina tissue culture as a source of indole alkaloids. Biopolym Cell 39(4):283-298.
8. Yadlovskyi OYe, Demchenko AM, Konvalyuk II, Mozhylevska LP, Bieda OA, Yarmoluk SM, Kunakh VA (2022) Study of the biomass of Rauwolfia serpentina Benth. tissue culture — a source of antiarrhythmic and hypotensive alkaloids — for acute and subacute toxicity. Pharmacology and Drug Toxicology 16(4):281-292.
9. Bieda OA, Konvaliuk II, Mozhylevska LP, Lukashov SS, Kunakh VA, Yarmoluk SM (2021) Determination of content of indole alkaloids in cell biomass of Rauwolfia serpentina Benth. ex Kurz tissue culture. Curr Issues Pharm Med Sci Pract 14(1):73-78.
10. Yadlovskyi OYe, Demchenko AM, Konvalyuk II, Mozhylevska LP, Bieda OA, Matiushok VI, Kunakh VA, Yarmoluk SM (2021) Study of the sedative effect of extract from tissue culture biomass of Rauwolfia serpentina. Farm Zh (2):78-86.
11. Monchak IL, Konvalyuk II, Balanda AO, Lukashov SS, Matiushok VI, Kunakh VA, Yarmoluk SM, Soloviev AI (2021) Study of the vasodilatory effect of extract from tissue culture biomass of Rauwolfia serpentina Benth. Farm Zh (3):71-80.