# Inductive launches Beacon-2 to turn drug properties into one dose estimate > Source: > Published: 2026-09-17 16:54:04+00:00 # Inductive launches Beacon-2 to turn drug properties into one dose estimate **Inductive's Josh Haimson and Ben Birnbaum are pairing the dose model with Indy, their new medicinal chemistry agent, inside live partner programs.** By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff) ยท Published Primary source: [PR Newswire](https://www.prnewswire.com/news-releases/inductive-launches-beacon-2-making-human-dose-computable-for-chemists-and-ai-agents-across-biopharma-302882274.html) ## Why it matters Beacon-2 moves Inductive beyond predicting isolated drug properties and into ranking compounds by a clinically meaningful endpoint. Its value will depend on prospective evidence that modeled dose estimates improve real laboratory and development decisions. [Josh Haimson](https://www.inductive.bio/team/josh-haimson?ref=runtimewire), co-founder and CEO of [Inductive](https://www.inductive.bio/?ref=runtimewire), launched Beacon-2 on September 17th, giving chemists and AI agents a system for estimating the efficacious human dose of a small molecule directly from its chemical structure. The product is the clearest expression yet of the bet Haimson and co-founder [Ben Birnbaum](https://www.inductive.bio/team/ben-birnbaum-ph-d?ref=runtimewire) have been building toward: drug discovery software becomes useful when it helps scientists make a consequential decision, rather than handing them another table of molecular properties to interpret. Haimson came to that problem through healthcare data. Before Inductive, he was director of product for Flatiron Health's machine-learning and data-curation organizations, where his groups worked with real-world evidence covering more than 2 million active cancer patients. He previously studied computer science at MIT and worked with Massachusetts General Hospital researchers on machine learning to predict how cardiac patients would respond to resynchronization therapy. Birnbaum, Inductive's president and CTO, was a senior director of engineering at Flatiron, leading machine-learning systems built around hundreds of millions of electronic health records. Earlier, he developed natural-language-understanding systems for Google Search and earned a computer science Ph.D. from the University of Washington. The founders have carried a product lesson from medical records into chemistry: a model's value depends on whether its output changes what a scientist does next. ### One number for the compound meeting Medicinal chemists typically compare compounds across potency and a collection of absorption, distribution, metabolism, excretion and toxicity measurements, known collectively as ADMET. Those properties interact. A potent molecule can still require an impractical dose if the body clears it too quickly, while a less potent compound can remain viable if its exposure and distribution are favorable. Beacon-2 predicts ADMET properties and potency, then feeds those estimates into mechanistic pharmacokinetic models. The output is a projected dose intended to help chemists decide which compounds deserve synthesis and laboratory testing. "Dose is what those properties all add up to," Haimson said in [Inductive's September 17th launch announcement](https://www.prnewswire.com/news-releases/inductive-launches-beacon-2-making-human-dose-computable-for-chemists-and-ai-agents-across-biopharma-302882274.html?ref=runtimewire). "Computing it from structure changes which compounds a team decides to make and helps them identify the highest quality compounds faster." That framing matters because medicinal chemistry groups already have access to a growing collection of property-prediction models. Inductive is trying to own the layer that converts their outputs into a progression decision. Beacon-2 sits inside [Compass](https://www.inductive.bio/solutions/compass?ref=runtimewire), Inductive's virtual lab platform, where the system is running on active partner programs, according to Inductive. Inductive says its [technical evaluation](http://www.inductive.bio/news/beacon-2?ref=runtimewire) covers projections across 20 drug programs and 325 public compounds from the ExpansionRx OpenADMET competition. The underlying Beacon ADMET models have also finished first in three consecutive blind challenges, according to Inductive. [OpenADMET's challenge archive](https://openadmet.org/blindchallenges/?ref=runtimewire) confirms that its recent PXR and ExpansionRx competitions attracted hundreds of participants and thousands of submissions. Those benchmark finishes test predictions on hidden experimental data. They do not establish that Beacon-2 can accurately forecast dose in human trials. The product combines several modeled quantities, and errors in potency, clearance, bioavailability or distribution can accumulate as they pass through the pharmacokinetic calculation. Prospective clinical comparisons will determine how well the estimates travel from a leaderboard to patients. ### The chemistry agent gets a target Beacon-2 also gives Haimson and Birnbaum a concrete objective for Inductive's agent strategy. Three days before the dose-model release, Inductive [introduced Indy](https://www.prnewswire.com/news-releases/inductive-bio-launches-indy-an-ai-chemistry-assistant-to-double-the-capacity-of-every-medicinal-chemist-302877082.html?ref=runtimewire), a medicinal chemistry agent designed to check assay data, analyze structure-activity relationships, propose analogs and manage synthesis queues. An agent that generates molecules still needs a way to rank them. Predicted dose provides that scoring function while compressing several competing properties into a single endpoint. Inductive tested the pairing by giving Indy the SARS-CoV-2 Mac1 inhibitor AVI-6451 and asking the agent to optimize it. Across five autonomous design cycles, Inductive says Indy produced IB-47, the best generated compound, with a 17-fold improvement in predicted human dose over AVI-6451. That figure describes movement in Inductive's model output, and the [technical post identifying both compounds](http://www.inductive.bio/news/beacon-2?ref=runtimewire) says the projected doses still need wet-lab validation. The demonstration still shows the operating model Haimson is selling. Indy proposes and evaluates chemical changes, Beacon-2 scores their effect on projected dose, and Compass places those recommendations in front of scientists running a drug program. Inductive's platform connects the virtual workflow to physical experiments through its synthesis operations and lab robotics. That integrated loop is where Inductive is trying to separate itself from vendors selling a standalone model or general-purpose chemistry chatbot. The founders want Inductive involved in the repeated design, synthesis and testing decisions that generate proprietary program data and improve partner-specific models over time. ### Investors funded the move from models to programs Inductive has announced about $29.3 million in equity financing. Haimson and Birnbaum [emerged from stealth in December 2023](https://www.inductive.bio/news/inductive-bio-emerges-from-stealth?ref=runtimewire) with a $4.3 million seed round co-led by Andreessen Horowitz Bio + Health and Lux Capital, with Character, Bessemer Venture Partners and AlleyCorp participating. In May 2025, Inductive [raised a $25 million Series A](https://www.inductive.bio/news/inductive-bio-raises-25m-series-a?ref=runtimewire) led by Obvious Ventures. Andreessen Horowitz Bio + Health, Lux Capital, S32, Character and Amino Collective joined the round, alongside angel investors including Oren Etzioni, Jeff Hammerbacher, Malay Gandhi and Jakob Uszkoreit. Inductive said the capital would fund model research, expansion of its shared data consortium and wider deployment of Compass. Beacon-2 puts those investments behind a sharper commercial proposition. Inductive says it supports more than 100 discovery programs, though it does not name the programs, disclose pricing or break out how many are paying deployments. Human-dose prediction also raises the standard of evidence customers should demand. A model that influences which compounds enter the lab can save synthesis cycles when it is right and discard promising chemistry when it is wrong. Haimson's product thesis is straightforward: chemists need an answer that matches the decision they are already trying to make. Beacon-2 gives them one number. Inductive's next task is showing how often that number survives contact with the lab and, eventually, the clinic.