Qure AI

Generated molecule is 1.98x more stable (lower energy)
Drug created by Machine Learning Model
An Existing Drug for Meningitis
Qure AI

Inspiration

Antibiotic resistance is rapidly emerging as one of the greatest global health crises. Yet, traditional drug discovery remains painfully slow, costly, and inefficient. We asked ourselves: what if artificial intelligence could accelerate the search for new drugs, and quantum computing could confirm which ones truly work? That question gave rise to Qure AI, a platform where AI predicts promising molecules, and Quantum validates their stability.

What it does

Qure AI is a drug discovery engine that combines artificial intelligence and quantum computing.
For this hackathon, we focused on meningitis related pathogens, using real bacterial datasets as our starting point.
It starts by studying real bacterial lab data along with the chemical features of molecules.
From this, the system learns what patterns make a molecule effective against dangerous pathogens.
Using that knowledge, the AI generates new molecules by mutating existing chemical structures.
After the AI proposes candidates, Qure AI takes an extra step: running quantum simulations (VQE) to test if those molecules are stable and realistic.
This two-step approach means we don't just create molecules that look effective on paper, we also confirm their stability and real-world viability.
While our project is built around meningitis, the same pipeline can scale to other diseases and pathogens, offering a faster and smarter way to discover new antibiotics.

How we built it

We began by collecting real bacterial datasets focused on meningitis pathogens like Neisseria meningitidis and Haemophilus influenzae.
The raw data (MIC, IC50, IZ endpoints, etc.) were cleaned and standardized so the system could learn consistent patterns.
Using RDKit, we created a generator that mutates SMILES strings to produce brand-new candidate molecules.
Each generated molecule was then passed to our machine learning model (Support Vector Machine), which predicted how effective it might be against meningitis bacteria.
The most promising candidates were translated into a Hamiltonian, a mathematical model of their quantum energy.
We used the Variational Quantum Eigensolver (VQE) to find the lowest energy states, checking whether those molecules would be stable in reality.

Challenges we ran into

Cleaning and normalizing lab data like MIC, IC50, and IZ endpoints was tricky because sometimes a lower value meant stronger activity, while in other cases a higher value was better.
Making RDKit mutations realistic, many random changes produced molecules were chemically invalid so we had to filter carefully.
Running the Variational Quantum Eigensolver (VQE) with active space reduction and time costly optimization. Quantum simulations are powerful but also resource-intensive, especially in a hackathon setting.

Accomplishments that we're proud of

Learned how to design and connect multiple advanced tools, and built a working end-to-end pipeline where new molecules are generated, scored by AI, and validated with quantum computing.
Gained experience training SVM models on real meningitis datasets and successfully applied them to predict molecule effectiveness.
Explored quantum computing in practice by implementing the Variational Quantum Eigensolver (VQE), translating molecules into Hamiltonians, and testing their stability.

What we learned

How to read and prepare biological assay data for machine learning.
How to generate valid molecules using RDKit.
How to train SVM models for drug prediction. = How to apply quantum computing (VQE + Hamiltonians) to test stability.

What's next for Qure AI

Right now, Qure AI is focused on meningitis, using real pathogen datasets. Our next step is to expand the pipeline to other diseases and pathogens, scaling the approach so it can accelerate drug discovery across a much wider range of global health challenges.