# SandboxAQ launches AQPotency to screen drugs without solved protein structures > Source: > Published: 2026-08-19 16:41:25+00:00 # SandboxAQ launches AQPotency to screen drugs without solved protein structures **Jack Hidary is putting SandboxAQ's numerical models behind Claude, with claimed pricing of $1 per 1,000 molecule-target comparisons.** By [RuntimeWire Staff](/author/runtimewire-staff) ยท Published Primary source: [PR Newswire](https://www.prnewswire.com/news-releases/sandboxaq-launches-aqpotency-bringing-ultrafast-virtual-screening-to-drug-discovery-without-a-solved-protein-structure-302855441.html) ## Why it matters AQPotency targets a costly early-stage bottleneck by screening compounds without a solved protein structure. SandboxAQ still needs public benchmarks showing that its speed and price produce better experimental hits. [SandboxAQ](https://www.sandboxaq.com/?ref=runtimewire) announced [AQPotency](https://www.sandboxaq.com/models/aqpotency?ref=runtimewire) on August 19, opening a virtual-screening model that ranks potential drug compounds without requiring a solved 3D structure of the protein being targeted. The general availability release, detailed in [SandboxAQ's announcement](https://www.prnewswire.com/news-releases/sandboxaq-launches-aqpotency-bringing-ultrafast-virtual-screening-to-drug-discovery-without-a-solved-protein-structure-302855441.html?ref=runtimewire), puts the model inside Claude through Anthropic's Model Context Protocol, or MCP. Researchers can also access AQPotency through SandboxAQ's website, with a Google Cloud Marketplace release planned to follow. [Jack Hidary](https://www.sandboxaq.com/company/leadership/jack-hidary?ref=runtimewire) formed SandboxAQ while leading AI and quantum-technology teams during six years at Google. His scientific background predates both businesses: he studied neuroscience at Columbia University and later worked on functional brain imaging and artificial neural networks as a Stanley Fellow at the National Institutes of Health. SandboxAQ spun out of Alphabet in 2022 around Hidary's thesis that AI paired with physics, chemistry and scientific data can produce useful numerical answers for industries where plausible language is beside the point. AQPotency is a direct test of that thesis. Hidary is also testing whether a conversational model can become the storefront for specialized scientific software, leaving Claude to interpret a researcher's request while SandboxAQ performs the quantitative work underneath. ### Screening without a crystal structure AQPotency estimates the potency of a molecule against a biological target, expressed by SandboxAQ as a predicted pIC50 value. That measurement helps researchers prioritize which compounds should proceed to laboratory testing. It does not establish that a compound will become an effective or safe drug. Many established virtual-screening workflows depend on a detailed representation of a protein's 3D structure. Those structures can come from experimental methods or computational predictions, but usable models are unavailable or unreliable for some targets. SandboxAQ says AQPotency can work from the molecule and target without that structural input, allowing researchers to rank candidates for programs that cannot begin with conventional docking. SandboxAQ says the model runs on ordinary computing hardware, returns results in seconds and costs as little as [$1 per 1,000 molecule-target comparisons](https://www.prnewswire.com/news-releases/sandboxaq-launches-aqpotency-bringing-ultrafast-virtual-screening-to-drug-discovery-without-a-solved-protein-structure-302855441.html?ref=runtimewire). Each result includes a confidence estimate and an indication of whether the target resembles the data range where AQPotency performs reliably. Those uncertainty measures could help researchers separate a usable ranking from an extrapolation that deserves more caution. The model can reverse the search as well. A researcher can start with a molecule and ask AQPotency to rank proteins that the molecule may affect. That workflow can support off-target screening or help investigate a compound that produces an observed biological effect through an uncertain mechanism. SandboxAQ's drug-discovery operation includes more than 70 specialists, including more than 48 domain PhDs, according to its [drug-discovery materials](https://www.sandboxaq.com/solutions/drug-discovery?ref=runtimewire). The group spans computational and medicinal chemistry, biology, bioinformatics, AI and cloud-scale molecular simulation. ### Claude becomes the distribution layer SandboxAQ first outlined its [Claude integration](https://www.sandboxaq.com/press/sandboxaq-integrates-its-quantitative-ai-models-with-anthropics-claude-via-mcp?ref=runtimewire) in May, when AQPotency remained on a waitlist. At the time, AQCat Adsorption Spin, a model for catalyst discovery, was the first SandboxAQ LQM accessible through Claude. AQPotency's release turns the drug-discovery portion of that roadmap into an available product. MCP reduces the amount of custom integration work needed to call SandboxAQ's models from a conversational interface. A scientist can describe a screening task in plain English rather than build infrastructure around an inference endpoint. The interface lowers deployment friction. The scientific value still rests on the underlying model, its training data and the quality of the experimental work used to confirm its predictions. The same distribution plan extends to [Google Cloud Marketplace](https://www.sandboxaq.com/press/sandboxaq-to-bring-large-quantitative-models-to-google-clouds-marketplace?ref=runtimewire). SandboxAQ announced in June that AQPotency would follow its catalyst model onto Google's marketplace. Selling through Claude and cloud marketplaces gives SandboxAQ access to research teams inside software environments they already use, a practical route for a scientific-model vendor seeking enterprise adoption. ### Experimental evidence remains the test SandboxAQ says AQPotency has produced ["experimentally validated impact" in eight customer programs](https://www.prnewswire.com/news-releases/sandboxaq-launches-aqpotency-bringing-ultrafast-virtual-screening-to-drug-discovery-without-a-solved-protein-structure-302855441.html?ref=runtimewire). SandboxAQ has not identified those eight programs or published model-wide benchmark results covering accuracy, calibration, false positives or prospective hit rates. The launch does include examples from Parkinson's disease research. [Dario R. Alessi](https://www.dundee.ac.uk/people/dario-alessi?ref=runtimewire), director of the University of Dundee's MRC Protein Phosphorylation Unit, said SandboxAQ's models helped researchers explore a larger biochemical space while improving activity and selectivity. Gary W. Miller of Columbia University's Mailman School of Public Health said work with SandboxAQ identified selective binders for SV2C, a membrane protein associated with dopamine signaling, from a commercial molecule library. Those collaborations provide evidence that the models have entered real research workflows. They do not provide enough public data to compare AQPotency's overall performance with other screening systems. That distinction matters in a field crowded with computational hits that fail to survive laboratory testing. A [2025 review of virtual screening](https://pubmed.ncbi.nlm.nih.gov/39862145/?ref=runtimewire) found a persistent gap between the volume of predicted hits and experimental validation, along with a lack of standardized evaluation criteria. AQPotency also enters a market where competitors are attacking the same cost and structure bottlenecks from different directions. [Schrodinger's Virtual Screening Web Service](https://www.schrodinger.com/platform/products/virtual-screening-web-service/?ref=runtimewire) uses 3D docking and machine learning to screen libraries exceeding one billion compounds, with Schrodinger advertising results in roughly one week. [Isomorphic Labs](https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier?ref=runtimewire) says its IsoDDE system can predict binding affinity and identify potential binding pockets from an amino-acid sequence, placing it closer to SandboxAQ's structure-free pitch. ### Hidary's broader quantitative AI bet AQPotency arrives after SandboxAQ raised more than $450 million in an [April 2025 Series E](https://www.sandboxaq.com/press/sandboxaq-closes-450m-series-e-round-with-expanded-investor-base?ref=runtimewire), taking disclosed funding since the Alphabet spinout above $950 million. Investors in the round included Google, NVIDIA, BNP Paribas, Ray Dalio and Horizon Kinetics, joining backers including Eric Schmidt, Breyer Capital, T. Rowe Price, Paladin Capital and S32. SandboxAQ has also moved deeper into government-backed industrial research. In June, [RuntimeWire reported that SandboxAQ secured a $500 million CHIPS research and development award](/article/jack-hidary-sandboxaq-500-million-chips-materials-award) for work on semiconductor materials, with the U.S. government receiving a minority stake under the agreement. The drug launch shows the commercial side of the same strategy. Hidary has assembled specialized scientific teams and raised enough capital to build proprietary quantitative models across several industries. Claude and Google Cloud provide distribution without requiring SandboxAQ to replace the software environments researchers already know. AQPotency's low claimed inference cost and structure-free design could widen the number of targets and molecule libraries that smaller research groups can examine. Adoption will depend on results that survive the wet lab. In drug discovery, an inexpensive prediction earns its value only when it helps scientists buy fewer compounds, run fewer assays and find better starting points.