DrugNet – Prognica Labs

DrugNet
Knowledge Graph

The world’s most connected biomedical knowledge network – 500M+ curated relationships across drugs, genes, proteins, diseases, and pathways, continuously updated in real time.

ChEMBL 2.4M compounds
DrugBank 14,000+ drugs
UniProt 570K reviewed proteins
OMIM 27,000+ genetic disorders
PubMed 36M publications
ClinicalTrials.gov 480K trials
DisGeNET 1.2M disease-gene associations
STRING 20B protein interactions
KEGG 560 metabolic pathways
Reactome 15K reactions
BindingDB 2.8M binding measurements
TCGA 33 cancer types

What is DrugNet

The intelligence layer beneath every repurposing discovery

Traditional drug databases are isolated silos. DrugNet is the connective tissue – a living, heterogeneous knowledge graph that links every drug, target, gene, disease, pathway, and clinical observation into a single, queryable intelligence network. When you ask whether a failed SGLT2 inhibitor might treat heart failure, DrugNet draws on 500 million relationships to answer you in seconds, not months.

DrugNet is built on a property graph model with five primary entity classes and over 40 relationship types, enabling multi-hop traversals that surface connections invisible to any individual database.

Multi-Hop Repurposing Queries

Traverse from a drug compound through targets, pathways, and genetic associations to surface disease indications where mechanistic evidence already exists — but no therapeutic hypothesis has been formed.

Real-Time Evidence Updates

DrugNet ingests and processes new PubMed publications within 48 hours using our NLP pipeline, ensuring the graph reflects the current state of biomedical science – not a static snapshot from years ago.

Pathway & Mechanism Tracing

Follow a drug’s downstream mechanism across signalling pathways, gene regulatory networks, and protein interaction cascades. Understand not just what a drug hits, but what that hit sets in motion biologically.

Evidence Scoring & Confidence

Every relationship in DrugNet carries a scored evidence trail – number of supporting publications, experimental validation level, replication count, and source reliability. Decisions are grounded, not guessed.

Cross-Database Deduplication

When the same compound appears under 12 different names across ChEMBL, DrugBank, and PubChem, DrugNet™ resolves them to a single canonical node using InChIKey matching and ML-based entity resolution.

API-First, Integration-Ready

Full REST API and Python/R SDK. Query with SMILES strings, gene symbols, disease MONDO codes, or free text. FHIR R4-compatible outputs for direct integration with clinical informatics systems and EHRs.

Case Studies

From graph query to
clinical hypothesis

Oncology × Cardiology

SGLT2 Inhibitor Repositioned for Heart Failure with Reduced Ejection Fraction

Helped a small-size pharma client to evaluate their shelved SGLT2 inhibitor, originally developed for Type 2 diabetes but discontinued after failing a superiority trial against empagliflozin. Identified a convergent mechanism via AMPK activation, NHE1 inhibition, and cardiac fibrosis suppression that had only been characterised in post-2018 literature – after the drug’s development had been halted. The graph surfaced 14 independent mechanistic pathways converging on cardiac protection.

Neurology × Immunology

Failed JAK Inhibitor Identified as Candidate for ALS Neuroprotection

Helped an academic spinout from a Austria to investigate whether any approved anti-inflammatory compounds shared mechanistic overlap with recently characterised TDP-43 aggregation pathways in ALS. Identified a discontinued JAK1/2 inhibitor whose off-target binding profile had been documented piecemeal across independent studies but never connected into a coherent ALS hypothesis.

Infectious Disease

Repurposed Antibiotic Compound Demonstrates Activity Against Drug-Resistant M. tuberculosis Strain

A health NGO in India tasked Prognica with identifying candidates from the existing antibiotics library that might exhibit activity against a novel drug-resistant TB strain. We queried for compounds with documented activity against related bacterial cell-wall synthesis enzymes, cross-referencing with genetic variant data from the resistant strain. The graph identified a compound originally developed for Gram-positive bacterial infections.