AI-Augmented Discovery of Natural Compounds for Multi-System Disease and Environmental Health: A Physiology-First Translational Framework

Abstract

Natural products represent a vast and underutilized source of therapeutic candidates across multiple disease domains. However, traditional discovery approaches often prioritize isolated potency or target affinity without sufficient consideration of human physiology, delivery constraints, or translational feasibility.

We present a physiology-first AI framework for natural compound discovery that integrates multi-source data, multi-model ranking, and human-guided arbitration to prioritize candidates with real-world therapeutic potential. The platform incorporates chemical databases, literature-derived signals, biological pathway mapping, and delivery-aware scoring to identify compounds with multi-system relevance. Early proof-of-concept work using natural product datasets demonstrates the ability to prioritize candidates with anti-inflammatory, antioxidant, and systems-level effects. Applications extend across neurodegenerative disease, cardiometabolic disorders, oncology, longevity, and environmental health.

As an illustrative use case, the platform supports identification of interaction-mediated biologics targeting microplastic and nanoplastic burden, including polysaccharides and enzyme classes evaluated through ex vivo experiments. This framework provides a scalable pathway for translating natural compound discovery into clinically relevant therapeutic strategies. This paper is also available at: https://doi.org/10.5281/zenodo.19752369

Figure 1. Physiology-First Versus Traditional Drug Discovery Approaches. Comparison of conventional pipelines that prioritize target-specific potency with a physiology-first framework that incorporates multi-system relevance, delivery feasibility, safety considerations, and translational potential earlier in the discovery process.

1. Introduction

Natural compounds have historically served as a foundation for therapeutic development, yet systematic exploration of their full potential remains incomplete. The combinatorial complexity of natural product space, combined with limited prioritization frameworks, has constrained translation into clinical use.

Traditional drug discovery pipelines often emphasize:

• single-target potency

• high-throughput screening without physiological context

• late-stage consideration of delivery and tolerability

• limited integration across biological systems

These limitations may reduce translational success rates.

There is increasing need for discovery systems that prioritize:

• human physiological relevance

• multi-system effects

• delivery feasibility

• safety and tolerability

• real-world implementation potential

2. Platform Architecture

The platform integrates heterogeneous data sources into a unified discovery framework.

Data Inputs

• natural product databases (e.g., COCONUT and related repositories)

• biochemical databases (UniProt, BRENDA, KEGG)

• compound activity databases (ChEMBL, PubChem)

• scientific literature (NLP-derived features)

• pathway and systems biology datasets

• internal experimental observations

Model Components

• graph-based models for compound–target relationships

• natural language processing for literature signal extraction

• ranking models for multi-objective prioritization

• multimodal integration pipelines

• human-in-the-loop arbitration layers

The platform is designed to operate across multiple disease domains without restriction to a single dataset or therapeutic area.

Figure 2. AI-Augmented Natural Compound Discovery Platform Architecture. Conceptual overview of a multi-source, multi-model system integrating chemical databases, literature-derived signals, and biological pathway information to generate and prioritize therapeutic candidates. A human-in-the-loop layer supports final arbitration and translational refinement.

3. Physiology-First Prioritization

A central feature of the platform is physiology-first ranking.

Rather than prioritizing potency alone, candidates are evaluated based on:

• systemic biological relevance

• multi-pathway modulation

• predicted safety and tolerability

• delivery feasibility (oral, inhaled, injectable, etc.)

• stability and formulation compatibility

• potential for combination therapy

This approach aligns discovery outputs with downstream translational requirements earlier in the pipeline.

4. Proof-of-Concept Embodiments

Initial proof-of-concept work has included:

• screening of natural compound libraries (e.g., COCONUT-derived subsets)

• prioritization of phytochemical classes such as polyphenols, flavonoids, and glycosides

• focused analysis of plant-derived compounds (e.g., hibiscus-associated compounds)

• identification of candidates with overlapping anti-inflammatory and antioxidant profiles

These studies demonstrate the feasibility of narrowing large chemical spaces into biologically relevant candidate sets.

5. Environmental Health Use Case: Microplastics and Nanoplastics

One illustrative application of the platform is the identification of compounds that may mitigate the biological burden of microplastics and nanoplastics. Rather than focusing exclusively on polymer degradation, the framework prioritizes interaction-mediated biologics that support sequestration, aggregation, barrier protection, and reduction of biologically active particle interfaces.

Early concept development has included polysaccharides, natural binding materials, and selected enzyme classes evaluated through ex vivo experiments. These approaches aim to promote clustering or neutralization of particulate matter, which may improve clearance potential or reduce accessible reactive surface area. This use case highlights the flexibility of the platform in addressing emerging environmental health challenges beyond traditional pharmacologic targets.

6. Translational Roadmap

The platform supports a staged translational pathway:

1. AI-based candidate generation and ranking

2. in vitro and ex vivo validation

3. formulation and delivery optimization

4. preclinical evaluation in disease-relevant models

5. integration with biomonitoring and real-world data systems

This pipeline is adaptable across therapeutic domains and compound classes.

7. Broader Applications

This framework and approach may be applied to:

• neurodegenerative disease (e.g., Dementia, Parkinson’s disease)

• cardiometabolic disease

• oncology

• inflammatory and immune-mediated disorders

• environmental toxicology

• longevity and resilience medicine

Figure 3. Broad Applications of AI-Driven Natural Compound Discovery. Illustration of how a unified discovery platform can generate candidate therapeutics across diverse domains including neurodegeneration, oncology, cardiometabolic disease, environmental toxicology, and longevity-focused interventions.

The platform is not restricted to any single disease or compound class and is designed for extensibility.

Conclusion

AI-augmented natural compound discovery represents an opportunity to bridge large-scale chemical diversity with clinically relevant therapeutic development. By integrating multi-source data, physiology-first prioritization, and translational design principles, the Intellicite Labs framework provides a scalable approach to identifying candidates with real-world potential. The inclusion of emerging applications such as environmental health further expands the scope of this platform, supporting a unified strategy for multi-system disease and resilience-focused therapeutics.

Keywords

natural compounds; AI drug discovery; translational medicine; physiology-first; multi-system therapeutics; environmental health; microplastics; nanoplastics; biologics; ex vivo validation

Representative Citations

1. Sorokina M, et al. COCONUT online: Collection of Open Natural Products. J Cheminform. 2021.

2. Chen Y, et al. Polystyrene nanoparticles modulate tumor microenvironment and α-synuclein-related pathways. J Hazard Mater. 2024.

3. Gou X, et al. Impact of nanoplastics on Alzheimer’s disease: Enhanced amyloid-β peptide aggregation and augmented neurotoxicity. J Hazard Mater. 2024.

4. Windheim J, et al. Micro- and Nanoplastics’ Effects on Protein Folding and Amyloidosis. Int J Mol Sci. 2022.

5. Yoshida S, et al. A bacterium that degrades and assimilates poly(ethylene terephthalate). Science. 2016. (PETase reference)

6. Li S, et al. Microplastics induce BBB disruption and neuroinflammation in rodent models. Environ Pollut. 2022.

7. Cui Y, et al. PET microplastics and endocrine/neuroinflammatory signaling. Environ Sci Technol. 2023.

8. Park JH, et al. Polypropylene microplastics enhance metastasis-related gene expression in human breast cancer cells. Environ Pollut. 2023.

9. Herrala M, et al. Microplastic exposure and endocrine disruption in male reproductive tissues. Environ Int. 2023.

10. Ding Y, et al. Nanoplastics exacerbate oxidative stress and mitochondrial dysfunction in neural cells. Environ Sci Technol. 2023.