SandboxAQ AI Skips Key Step to Accelerate Drug Discovery

PALO ALTO, CA – January 20, 2026 – Tech firm SandboxAQ today released AQAffinity, a powerful artificial intelligence model that promises to significantly accelerate early-stage drug discovery by bypassing one of its most stubborn bottlenecks: the need for a protein’s 3D structure. By predicting a drug candidate's potency based on sequence data alone, the tool aims to help scientists find promising new medicines faster and more cheaply.

Developed by the Alphabet spin-out, AQAffinity is built to work seamlessly with OpenFold3, a state-of-the-art, open-source model for predicting the structure of biomolecular complexes. The new tool essentially allows researchers to screen potential drugs against their protein targets without first having to solve the target's complex three-dimensional shape through costly and time-consuming laboratory experiments. This capability could fundamentally alter the risk-reward calculation for pharmaceutical companies, particularly when pursuing novel therapies for diseases linked to less-understood proteins.

De-risking the Discovery Pipeline

The journey of a new drug from lab to clinic is notoriously long, expensive, and fraught with failure. A major hurdle in the initial phase is identifying a 'hit'—a small molecule that effectively binds to a target protein involved in a disease. Traditionally, this process relies heavily on structure-based drug design, which requires a high-resolution 3D map of the target protein. Obtaining this map using methods like X-ray crystallography or cryo-electron microscopy can take months or even years, and for many proteins, it remains impossible.

AQAffinity sidesteps this entire process. By learning the complex patterns of molecular interaction directly from sequence data, it provides rapid predictions of binding affinity. This allows research teams to quickly triage vast libraries of potential drug compounds, letting them “fail fast” on unpromising candidates and focus resources on those with the highest probability of success. This is especially critical for de-risking challenging targets that have historically been considered 'undruggable' due to a lack of structural information.

"AI-driven binding affinity prediction has evolved from a niche method to a core component of modern drug discovery workflows," said Nadia Harhen, SandboxAQ's General Manager for AI Simulation, in a statement. She noted that the new model enhances OpenFold3 with a critical feature, giving it parity with competing platforms while adding new benefits for biopharma researchers.

An Open-Source Challenge to 'Black Box' AI

The launch of AQAffinity is not just a technical milestone; it’s a strategic move that champions a more open and collaborative approach to science. The AI drug discovery market is crowded with companies offering powerful, but often proprietary and opaque, platforms. These 'black box' systems can be difficult for external scientists to validate or customize, creating a dependency on a single vendor.

In stark contrast, SandboxAQ has released AQAffinity under the permissive Apache 2.0 license, making it and its underlying OpenFold3 model fully open-source and available on the popular AI repository Hugging Face. This transparency allows any organization to inspect the model's architecture, training data, and methods. Researchers can freely evaluate its performance using their own internal benchmarks, fine-tune it for specific targets, and integrate it into their workflows without costly, long-term commitments.

This open approach has been praised by leaders in the field, including Professor Mohammed AlQuraishi, the principal investigator at OpenFold. "With AQAffinity, SandboxAQ delivered critical affinity prediction capabilities to OpenFold3 within a few short weeks after the model's launch, showcasing the company's technical expertise and commitment to advancing AI-powered open-source tools," AlQuraishi stated. He emphasized that the OpenFold Consortium is dedicated to building an entire ecosystem of open tools to help accelerate drug pipelines.

From Deep Tech Vision to Practical Application

For SandboxAQ, which emerged from Alphabet Inc. in 2022 with a reported $500 million in funding, AQAffinity is a tangible demonstration of its ambitious vision. The company operates at the intersection of AI and quantum technologies, developing what it calls Large Quantitative Models (LQMs) to tackle complex problems in sectors ranging from life sciences to navigation and finance. Backed by a roster of high-profile investors including T. Rowe Price, Eric Schmidt, and Ray Dalio, the company is under pressure to translate its deep-tech expertise into real-world impact.

AQAffinity fits perfectly into this strategy. It represents a practical application of advanced AI that addresses a clear and significant market need. By leveraging the foundational work of the OpenFold Consortium, SandboxAQ has been able to rapidly develop and deploy a tool that not only showcases its technical capabilities but also aligns with a broader industry shift toward more efficient, computationally driven research and development.

Early industry feedback appears positive. OpenFold Consortium members began beta testing the model in December 2025. Sherry Liu, Co-founder and CTO at TamarindBio, a company that provides computational drug discovery services, commented on its potential. "AQAffinity will complement OpenFold3 with valuable binding affinity predictions from SandboxAQ," Liu said. "AQAffinity will add to Tamarind's toolkit to help our customers streamline virtual screening, improve hit-to-lead optimization, and accelerate small molecule discovery and design."

By providing this powerful capability as an open-source tool, SandboxAQ and the OpenFold Consortium are not just offering a new piece of software; they are fostering a transparent, collaborative ecosystem designed to accelerate the development of novel therapies and ultimately bring them to market.