deepmirror, in collaboration with Google Cloud, enabled Curie.Bio to run advanced protein structure models to tackle challenging, undrugged targets.

AI-powered protein structure prediction is revolutionizing drug discovery. Cutting-edge models like AlphaFold 3, Chai-1, and Boltz-1 allow scientists to predict protein complexes with unprecedented accuracy. However, applying these models to challenging targets requires overcoming significant computational and security barriers.

Curie.Bio, an innovative biotech venture fund and R&D platform, sought to leverage these powerful tools to enable hypothesis driven design for targets that have not yet been structurally enabled. Yet, deploying these models efficiently and securely proved to be a major hurdle due to:

• High Computational Costs – Running advanced protein structure models demands specialized GPUs, which are expensive and complex to maintain.
• Security Risks – Sending proprietary structural data through public APIs exposes valuable intellectual property to potential breaches.
• Limited Commercial Options – Existing solutions either lack robust security measures or fail to provide the necessary computational performance.

Curie.Bio’s initial attempts to self-host these models using open-source repositories quickly ran into roadblocks, including infrastructure limitations, the need for highly specialized hardware, and unacceptable data security risks.

deepmirror's Secure Structure Prediction Platform

To address these challenges, deepmirror, in close collaboration with Google Cloud, developed a secure structure prediction platform powered by specialized A100 GPUs. This solution was designed to provide a private, high-performance computing environment that safeguards sensitive structural data through deepmirror’s authentication systems and encrypted data transfer protocols.

Key Features of deepmirror’s Secure Platform:

• Private, Encrypted Computing Environment – Ensures proprietary data remains secure.
• Authenticated API – Enables users to safely upload and process structural data without relying on insecure open-source tools.
• Scalable High-Performance Computing – Provides seamless access to specialized GPUs for large-scale predictions.

By integrating these capabilities, deepmirror allowed Curie.Bio’s scientists to focus on their discoveries rather than infrastructure management. The platform’s secure API offered authenticated endpoints for submitting structures and receiving predictions while maintaining strict access controls to protect sensitive data.

Transforming Drug Discovery with Secure AI-Driven Predictions

With deepmirror’s secure structure prediction platform, Curie.Bio achieved:

• Running structure predictions efficiently without costly on-prem infrastructure.
• Full Control over Intellectual property - Eliminating security risks associated with public APIs.
• Generating hypotheses that drive the design of molecules in collaboration with portfolio companies to identify promising lead candidates.

“Structure enablement is a key inflection point for many drug discovery programs. However, getting to this point is not trivial, even with experimental evidence of ligand binding. deepmirror quickly enabled us to leverage empirical data in conjunction with AI structure prediction models to develop testable design hypotheses for our lead compounds.”

— Michelle Southey, Curie.Bio

As the industry moves toward increasingly complex and undrugged protein targets, secure, high-performance AI platforms like deepmirror’s will be critical for accelerating breakthroughs in drug discovery, empowering researchers to push the boundaries of what’s possible.

About deepmirror

deepmirror combines human ingenuity with AI to help biopharma teams focus on the most promising drug candidates so they spend less time on guesswork and more time impacting patient lives. deepmirror launched in late 2023, has raised more than $3M, and is trusted by teams across the globe.