New York, New York, September 17, 2026
Inductive Bio has launched Beacon-2, a new artificial intelligence system designed to predict the efficacious human dose of small-molecule compounds directly from chemical structure. The company says the technology enables chemists and AI agents to evaluate potential drug candidates before compounds are synthesized and tested in the laboratory. Beacon-2 combines predictions of absorption, distribution, metabolism, excretion and toxicity (ADMET) with estimates of molecular potency and mechanistic pharmacokinetic models to calculate an expected human dose. The launch represents an effort to make human-dose prediction a computable objective within preclinical drug discovery and AI-driven chemical optimization.
Beacon-2 Integrates Potency and ADMET Data
A major challenge in small-molecule drug discovery is determining whether a compound can achieve sufficient therapeutic activity at a dose that is practical and tolerable. Scientists traditionally evaluate multiple properties independently, including potency, pharmacokinetics, permeability, metabolic stability and toxicity. Beacon-2 is designed to bring these factors together into a unified prediction of efficacious human dose. According to Inductive, the system includes an ADMET-PK module that predicts chemical and pharmacokinetic properties, a potency module that estimates compound activity, and a PK/PD module that combines these outputs through physiologically informed pharmacokinetic models. This approach allows researchers to evaluate compounds based on their predicted overall therapeutic potential rather than optimizing individual molecular characteristics in isolation.
The company reports that the ADMET models supporting Beacon-2 have placed first in three consecutive OpenADMET blind challenges, competing against more than 750 entries from AI and pharmaceutical organizations. Inductive also says the technology was evaluated across 20 real-world drug programs and 325 publicly available compounds associated with the ExpansionRx OpenADMET competition. These evaluations are intended to assess how accurately the system can project properties relevant to human dose before extensive laboratory experimentation.
AI Agents Can Optimize Toward Human Dose
Advertisement
Beacon-2 is also designed to provide an objective that can be used by AI agents for autonomous compound optimization. Inductive tested the system with its medicinal chemistry agent Indy, using a recently disclosed SARS-CoV-2 compound as the starting point. Across five autonomous design cycles, Inductive reported that Indy improved the compound’s predicted human dose by 17-fold while optimizing the molecular structure with Beacon-2. The company argues that this type of computational feedback can help medicinal chemists identify promising compounds earlier in a discovery program. Instead of waiting for synthesis and experimental testing before assessing a compound’s likely dose characteristics, researchers can use in-silico predictions to prioritize molecules for laboratory work. The approach could potentially reduce the number of compounds requiring synthesis and accelerate the iterative cycle between molecular design, prediction and experimental validation.
Inductive’s broader Compass platform brings Beacon-2 together with other computational and chemistry capabilities. The company says Beacon-2 is already running on live drug-discovery programs with partners, while its AI systems and expert teams support more than 100 discovery programs with biopharma partners.
Beacon-2 Expands AI-Driven Drug Development
The launch reflects the growing use of artificial intelligence in preclinical drug discovery, where computational models are increasingly being used to predict molecular properties and guide medicinal chemistry decisions. Beacon-2 extends this approach by making predicted human dose the central optimization objective rather than focusing exclusively on potency or individual ADMET characteristics. Inductive describes efficacious human dose as a useful integrated measure because it reflects the relationship between drug potency and pharmacokinetic behavior. A lower predicted efficacious dose can also be relevant to considerations such as exposure and potential off-target toxicity, although computational predictions cannot replace laboratory, preclinical or clinical evaluation.
Importantly, Beacon-2 provides predictions rather than clinical measurements. Human dose, safety and efficacy ultimately require experimental and clinical validation. The system is intended to help researchers prioritize and optimize compounds earlier in the development process, not to replace the studies required before a drug can enter clinical use. For cGxP.wire readers, the launch highlights the convergence of AI, medicinal chemistry, ADMET modeling and pharmacokinetics. By attempting to make human-dose prediction computable directly from chemical structure, Beacon-2 represents another development in the application of AI to small-molecule drug discovery.
Source: Inductive Bio press release
