AI-Driven Network Pharmacology — Working Prototype
AI-NP is a functioning prototype — a trained PyTorch/PyTorch Geometric model, a live knowledge graph, and an interactive dashboard — mapping compound–target–pathway relationships inside multi-compound formulas to identify licensable, scientifically validated candidates for pharmaceutical partners.
The Platform
AI-NP is an AI drug-discovery and network-pharmacology platform, already implemented as a working prototype. Its core IP is the modeling engine — a pipeline that ingests formula composition data, cross-references bioactive compound and target-interaction databases, and ranks hypotheses for mechanism of action and candidate viability. Traditional Chinese Medicine formulas serve as its first, richly-documented training dataset.
A 1,000-herb and mineral materia medica archive, drawn from 16 generations of documented clinical practice.
Compound → target → pathway prediction, validated against known drug-target relationships.
Ranked, validated leads — licensed or co-developed with pharmaceutical manufacturing partners.
Regulatory approach: AI-NP identifies and validates drug candidates for licensing — clinical development, regulatory approval, and manufacturing are carried out by our pharmaceutical partners, not by AI-NP directly.
Prototype Status
AI-NP has moved from idea to working software. The core pipeline is built, trained, and running end-to-end on a live dashboard, with real graph data, a trained deep-learning model, and reproducible experiment tracking.
Herb → Compound → Target relationships seeded from TCM reference databases (TCMSP, TCMID, SymMap), queryable through an interactive network visualization.
A drug-target interaction model trained end-to-end, with training loss reduced from 21.98 to 2.86 MSE across 5 epochs, tracked and versioned in MLflow.
A GNN layer (GATConv graph-attention) modeling the network directly — computing per-edge attention weights and multi-hop message passing across the compound-target graph.
RDKit + AutoDock Vina integration for 3D structural validation of predicted compound-target interactions.
A composite Synergy Index quantifying target overlap and pathway convergence across a formula's constituent compounds — demonstrated on Ginseng at a Synergy Index of 82%.
A live, bilingual (EN/中文) dashboard for herb search, interactive network graphs, and real-time GNN simulation — the interface partners and reviewers can see the platform run.
Market Validation
Pharmaceutical companies are paying substantial sums to license AI-discovered drug candidates without building or acquiring the underlying AI platform themselves — directly validating the discovery-and-license model AI-NP is built on.
McKinsey estimates AI, particularly generative AI, could contribute this much annual value to pharma — primarily by accelerating discovery and development.
Insilico Medicine's research collaboration with Sanofi, on $21.5M upfront — a comparable AI-driven drug discovery licensing structure.
Insilico's upfront exclusive license to Exelixis and cardiometabolic collaboration with Qilu Pharmaceutical, respectively.
Team
16th-generation TCM practitioner; clinics across two EU countries; author of the platform's founding formula archive.
Develops the core platform and user-facing tools, and supports integration between data pipelines and modeling components.
Leads company strategy, fundraising, and core platform development.
Open Roles — Hiring with Program Funding
Will define the biological hypotheses driving the platform and validation criteria for model outputs.
Will build and statistically validate the predictive models under the Pharmacologist's direction.