Introduction: What Social Networks and Medicine Have in Common
It might sound surprising, but the same AI algorithms that suggest new LinkedIn connections could be the key to unlocking new uses for existing medicines. When LinkedIn recommends a distant relative or an old friend from a completely unrelated industry, it’s not magic — it’s math. Specifically, it’s powered by Graph Neural Networks (GNNs), a cutting-edge technology that’s now being adapted for use in biomedicine and drug discovery.
How LinkedIn’s Graph Model Works
On LinkedIn, your profile is a node in a giant network graph. The links (or edges) between you and your connections — plus the people they know, companies you follow, or posts you like — create a complex web of interactions. These algorithms don’t just consider your data, but also aggregate insights from your broader network. That’s how the platform knows you might know someone you’ve never even interacted with online.
Each layer of the graph adds context and hidden relationships. After a few iterations, the algorithm can make highly accurate predictions based on indirect, yet meaningful, connections.
Repurposing the Graph: From Social Media to Biomedicine
What does this have to do with drug repurposing? Quite a lot, actually. Creating a new drug is notoriously expensive and time-consuming. That’s why scientists are turning toward drug repurposing, or finding new uses for existing, already approved medications.
Instead of building new compounds from scratch, researchers now create graph models that connect drugs and the proteins they interact with. Each node represents a drug or protein, and each edge reflects a known interaction, just like LinkedIn’s people and connections.
The Power of Prediction in Drug Discovery
These graphs allow researchers to use AI to predict unknown interactions, giving scientists a data-driven shortlist of drugs that might work for diseases they weren’t originally intended to treat.
Databases like DrugBank — which now lists over 2,700 approved drugs — provide the raw material for these graphs. Once created, models can evaluate thousands of possible interactions within minutes, saving years of lab time and millions in research funding.
Meet GeNNius: The Network Changing the Game
At the University of Navarra, researchers at the Computational Biology and Translational Genomics lab have built GeNNius, an AI model inspired by these same graph structures. It’s designed to process drug-protein networks quickly and accurately, achieving over 23,000 evaluations per minute.
While GeNNius still faces challenges, particularly when working with unknown or poorly documented molecules, its speed and predictive power represent a major leap forward in the field. The ultimate goal? To evolve these models into tools that provide personalized treatment recommendations for individual patients based on their unique biological profile.
Conclusion: A Future Where AI Connects More Than People
LinkedIn’s algorithm might help you land your next job — but the underlying technology could also help someone find a life-saving treatment faster. By borrowing AI models from social media platforms, scientists are building a smarter, faster way to discover new medical possibilities. With continued research and better datasets, tools like GeNNius could transform how we approach public health, making medicine more efficient, affordable, and personalized for all.
To dive deeper into how graph-based AI is transforming health science, explore DrugBank and the latest research at the University of Navarra.
