deepmirror™ enabled Tes Pharma, a leading biotech focused on delivering first-in-class drug therapies for diseases with high unmet medical need, to screen 1 billion molecules in just a few hours, unlocking novel chemotypes against a transcription factor, belonging to one of the toughest target classes in drug discovery.

Advances in Artificial Intelligence (AI) and Deep Learning (DL) enable drug hunters to build predictive models of protein-ligand binding affinity from small datasets of biological activity. However, screening ultra-large libraries with billions of compounds at scale remains a technical challenge because:

• it requires dedicated computational resources,
• even with the right hardware, this scale stretches the limits of most in-house software tools,
• most available tools lack robust reliability scoring, making it difficult to trust and prioritise hits at scale.

Tes Pharma was advancing a programme against a transcription factor, a protein class notoriously difficult to target because it often lacks clear binding pockets. The team was looking for a faster and more scalable way to explore diverse chemical space and uncover novel chemotypes. To solve this, they built a predictive model on deepmirror™ using their in-house biological activity data. Over 1 billion molecules from a vendor library were screened in just a few hours to predict their binding properties using deepmirror’s Application Programming Interface (API).

Dr. Francesco Greco, Head of Computational Chemistry at Tes Pharma said: “We were impressed with how user-friendly and efficient screening the full 1B library has been. Just a few hours is really fast.”

deepmirror™ orchestrated the entire screening by automatically scaling the infrastructure, processing hundreds of millions of molecules in parallel, and returning predictions with uncertainty quantification to guide prioritisation. This large-scale campaign highlighted multiple novel scaffolds predicted to modulate the transcription factor target, offering fresh hypotheses and promising directions for follow-up exploration.

Dr. Ryan Greenhalgh, CTO & Co-Founder at deepmirror™ commented: “We built deepmirror™ to make what was once aspirational - building a model and screening 1 billion compounds in a single day - standard practice with just a few lines of code. By abstracting away the infrastructure, our platform allows discovery teams to adopt AI-powered screening effortlessly and at scale.”

Whether you're working with generative chemistry, vendor libraries, or proprietary collections, deepmirror™ is ready to scale with you.

Try it yourself

Sign up for a deepmirror™ account at start screening at scale today with this ready-to-run notebook.