Inspiration
Drug discovery is expensive, slow, and involves screening huge numbers of compounds before researchers can identify promising candidates. We wanted to explore how AI and computational tools could help researchers narrow that search earlier, while still being transparent about uncertainty and limitations.
What it does
Fairuz is an AI-assisted molecular screening platform that helps prioritize potential antiviral compounds for further investigation. It works with real molecular structures, calculates key molecular properties, ranks compounds based on their screening potential, and explains why each compound received its score.
Fairuz does not claim that a compound will work. Instead, it helps answer a more responsible question: which compounds might be worth investigating next?
How we built it
We built Fairuz in Python using tools including RDKit for molecular processing and property calculation, with a Streamlit-based interface for interacting with the screening pipeline.
The system validates molecular structures, calculates descriptors such as molecular weight, LogP, TPSA, hydrogen-bond donors and acceptors, applies a transparent ranking system, and produces explanations showing which factors contributed to each compound's result.
We also designed the pipeline to clearly distinguish computational predictions from experimental evidence.
Challenges we ran into
One of the biggest challenges was working with biological and chemical data responsibly. Drug-related datasets can be incomplete, biased, or inconsistent, so we had to avoid presenting computational results as biological truth.
We also had to balance technical sophistication with explainability. A more complicated model is not necessarily better if researchers cannot understand why it produced a result.
Accomplishments that we're proud of
We are proud that Fairuz became more than just a prediction model. We built a complete screening workflow that combines real molecular data, molecular property analysis, ranking, explainability, validation, automated testing, and a usable interface.
We are especially proud of the transparency built into the system. Every ranking can be explained, and the platform clearly communicates its limitations instead of overclaiming what AI can do.
What we learned
We learned that responsible AI in healthcare is not only about model accuracy. Data quality, bias, uncertainty, explainability, and the way results are communicated are equally important.
We also gained experience working with cheminformatics tools, molecular data, scientific Python libraries, testing, and designing AI systems where transparency is a core feature rather than an afterthought.
What's next for Fairuz
The next step would be to expand Fairuz using larger and more diverse experimentally validated datasets, improve the ranking methodology, and evaluate its predictions against real laboratory results.
Longer term, Fairuz could incorporate additional stages of early drug discovery while keeping the same principle at its core: AI should support researchers' decisions, not replace scientific validation.
Built With
- chatgpt
- codex