Cambridge, MA — June 22, 2026 — Receptor.AI today announced the release of StratAI™, an agentic decision engine for drug discovery at the core of its PharmaSphere™ ecosystem.

StratAI™ runs the discovery programs Receptor.AI carries out with its pharma and biotech partners, working across all three of the company's modality platforms — small molecules, peptides, and biologics. It drives every decision in a discovery program — from program-level strategy down to individual experiments and molecules — and adapts the program as new evidence accumulates. For partner teams, this translates into fewer experimental cycles, lower program costs, and faster convergence to validated drug candidates.

StratAI™ is designed to address a problem that has become increasingly visible across the field. Over the past decade, target identification, structure prediction, generative chemistry, screening, and lab automation have all become dramatically faster and cheaper. Success rates have not followed. Programs run more experiments than ever, in less time, and still fail at roughly the same rate — because faster execution alone does not produce better decisions. The harder question is no longer how to execute a discovery program faster, but how to make the right decisions throughout it: strategic, experimental, and molecular alike.

“The hardest problem in drug discovery is no longer running experiments — it’s deciding what to do at every step,” said Dr. Alan Nafiiev, Founder & CEO of Receptor.AI. “We built StratAI™ because the bottleneck has moved from execution to judgment, and judgment under uncertainty is not something you solve by improving AI models. It requires a different kind of system, and that system is what now works at the core of every program we run with our partners.”

From Faster Labs to Smarter Decisions

AI drug discovery has made enormous progress over the past decade — faster experiments, better structure prediction, generative chemistry, and lab automation. But these advances have shifted the bottleneck rather than removed it. The hard question is no longer how to do more, but how to decide what to do at each level of a program — and most programs today fail because those decisions are made on instinct, in isolation, or too late.

Real progress requires a different kind of system: one that makes coherent decisions across every level of a program at once. Whether to pursue a target at all. Which modality to commit to. Where the biggest property risks sit, and how to design the assay cascade around them. Which experiment in the next batch will actually move the program forward. Which molecules deserve the next synthesis slot. These decisions are not independent — the ones above shape the constraints for the ones below, and the ones below produce the evidence that refines the ones above. Most platforms still treat them as separate predictions stitched together at the end. That is how wasted experiments accumulate.

How StratAI™ Works

StratAI™ is built as an adaptive decision engine with a hierarchical architecture. It operates across four decision levels that mirror how drug discovery programs actually run — program, campaign, experiment, and molecule — supported by an integrated execution layer that connects computational and experimental methods, and a closed-loop learning system that refines the entire stack with every new result.

Every program begins with two kinds of input:

Program context

Each program starts from a specific scientific situation defined by the partner team: the target and its biology, the modality being pursued, the disease indication, the success criteria, the budget and timeline, and the constraints that matter — IP, safety, regulatory. This context anchors every downstream decision to the program at hand.

Knowledge and data sources

Alongside the program's own context, StratAI™ has access to a continuously growing base of resources: internal data accumulated across Receptor.AI programs (assays, compounds, studies), public scientific data (literature, patents, databases), the platform's library of AI/ML models and computational tools, and expert knowledge from internal and external scientists. New programs do not begin from a blank slate — they begin from everything Receptor.AI has already learned.

On top of these inputs, StratAI™ operates a four-level hierarchy of decisions:

Program-level decisions

At the program level, StratAI™ makes the highest-stakes strategic calls — those that determine whether a program proceeds and how it should be shaped. The system continuously estimates the program's probability of success (PoS) and re-evaluates it as new evidence accumulates. Key program-level decisions include:

• Go / No-Go — whether the program should advance, pivot, or be discontinued.
• Target prioritization — which targets warrant deeper investment.
• Modality strategy — whether to pursue a small molecule, peptide, or biologic approach.
• Portfolio & risk assessment — how this program fits against others in the partner's portfolio.

This level produces the program's strategy and plan, its explicit resource allocation, and the success criteria that anchor everything below.

Campaign-level decisions

A program is delivered through one or more campaigns — coordinated optimization efforts focused on a specific set of properties. At this level, StratAI™ defines the shape of the optimization itself: what properties matter most, where the program's biggest risks sit, and how the experimental cascade should be structured to surface those risks at the right time. Key campaign-level decisions include:

• Property priorities — which properties (e.g., affinity, selectivity, permeability, metabolic stability) the campaign will optimize.
• Bottleneck analysis — which property is most likely to fail the program and where focus should concentrate.
• Design space definition — the chemical, sequence, or structural space the campaign will explore.
• Assay cascade design — which assays run in which order, at which fidelity.

The result is an optimization strategy, the assay cascade itself, and the milestones that mark progress within the campaign.

Experiment-level decisions

Within each campaign, StratAI™ decides which experiments to run and in what order. This is where formal optimization theory matters most concretely: for every open question, the system computes the expected information gain of each available experiment relative to its cost, and selects those that maximize the campaign's rate of learning per unit of resource. This is sequential experimental design under uncertainty — the same mathematical foundation behind Bayesian optimization and acquisition-function-based active learning, applied here across an entire campaign. Key experiment-level decisions include:

• Experiment selection — which question to answer next, given what the campaign already knows.
• Multi-fidelity routing — whether a question is best answered by a fast computational screen, a higher-fidelity simulation, or a wet-lab assay.
• Batch composition — which compounds go together in the next experimental batch to maximize what the campaign learns.
• Stopping rules — when enough has been learned and the campaign should advance to the next phase.

What comes out is a concrete experiment plan, the expected information gain it should deliver, and the decision rationale — what each experiment is meant to resolve.

Molecule-level decisions

Finally, at the molecule level, StratAI™ decides which specific compounds advance — through ranking, design, and selection. This is multi-objective optimization in practice: candidates are evaluated across many competing objectives simultaneously — potency, selectivity, ADMET, synthesizability, IP — and the system surfaces those that perform best across the full set, rather than narrowly optimizing for one endpoint at the expense of others. Key molecule-level decisions include:

• Molecule prioritization — which candidates from the current pool are most worth advancing.
• Analog & series ranking — which chemical series or analog families deserve continued focus.
• De novo suggestions — when to generate new molecules outside the current pool.
• Selection for synthesis / testing — which molecules go into the next experimental round.

This level produces a ranked list of molecules with confidence scores, structural design suggestions, and synthesis priorities for the next round.

Integrated execution layer

The four decision levels above act on a single integrated execution layer that brings together computational and experimental methods. Computational methods include docking and scoring, molecular dynamics, ADMET / PK prediction, generative design, and retrosynthesis. Experimental methods span in vitro, cellular, biophysical, and in vivo PK assays, executed either in the partner's own laboratory or through a network of pre-integrated CRO partners. StratAI™ moves between them as the decision requires — calling a docking run, a molecular dynamics simulation, or a wet-lab assay depending on what the current question needs and how confident the system already is about the answer.

Closed-loop learning and adaptation

Every result that comes back from the execution layer — whether a computational prediction or a wet-lab readout — feeds back into the system. StratAI™ integrates the new data, analyzes how it compares to expectations, updates its picture of the program, and refines its strategy for the next cycle. A decision made at any level can be revised when new evidence calls for it: a campaign-level bottleneck can shift after an unexpected assay result; a program-level PoS estimate can be re-anchored after a key experiment lands. This is what makes the architecture adaptive — each cycle sharpens the system's calibration for the cycles that follow.

One Decision Engine Across Modalities and Labs

StratAI™ serves as the shared decision layer across all three platforms in PharmaSphere™ — small molecules, peptides, and biologics. The chemistry differs between these modalities, but the underlying decision problem does not. As a result, lessons learned in one modality strengthen decision-making in the others, giving partners the benefit of a continuously improving system regardless of which platform they engage.

The system is also designed to be lab-agnostic. StratAI™ does not require any specific laboratory infrastructure. It works with the partner's own laboratory or through a network of pre-integrated CRO partners, with experimental design and prioritization driven by StratAI™ and wet-lab results feeding back into the next cycle. Partners adopt the decision intelligence without committing to a particular hardware stack.

This architecture also means each new program benefits from every program that came before it. StratAI™ accumulates experience in which strategies work for which kinds of targets, so new partners are not starting from scratch — they are starting from a system that has already learned how to decide.

Case Study: Optimizing a Cyclic Peptide for an Intracellular PPI Target

In a recent program, StratAI™ was applied to optimize a 16-residue cyclic peptide targeting an intracellular protein–protein interaction involved in endosomal recycling. The starting peptide bound the target at 780 nM but was effectively impermeable to membranes (logPerm = −8), making intracellular delivery the limiting factor for the program.

The two objectives — improving binding affinity and improving membrane permeability — typically pull in opposite directions, since the hydrophobic residues that drive binding also tend to trap a peptide in the lipid bilayer. The space of possible modifications across 14 mutable positions, including non-canonical amino acids, exceeded one billion candidate sequences.

StratAI™ approached affinity and permeability as a single multi-objective problem rather than improving one parameter at a time. Fast AI models screened the sequence space broadly, and physics-based simulations were called in selectively — not on the top-ranked predictions, but on the candidates where AI-model uncertainty had accumulated and where a higher-fidelity readout would most sharpen the next round of predictions. This targeted use of physics-based methods is what shifts the final ranking from one driven by overconfident model outputs to one supported by the right depth of evaluation in the right places.

For wet-lab validation, the 20 top-predicted peptides were synthesized and tested. 9 of those 20 improved both affinity and permeability. The best candidate reached ~250 nM affinity and logPerm = −5.92 — a roughly three-fold affinity gain and over a hundred-fold improvement in passive permeability in a single round.

A second round of optimization explored amide-group shielding strategies, pushing predicted permeability further to logPerm = −5.63 while preserving the affinity gains already achieved. The same StratAI™ logic operates across small-molecule and biologics programs — the chemistry differs, but the decision problem and the system solving it remain the same.

Conclusion

For partners working with Receptor.AI, StratAI™ changes the practical shape of a discovery program. Pharma and biotech teams reach validated drug candidates faster, on optimized assay cascades, with significantly better resource and cost efficiency. Each program is structurally de-risked, because every decision is traceable, every assumption is testable, and the probability of success is continuously re-estimated as evidence accumulates. The molecules that emerge are higher quality — selected through multi-objective optimization across the full set of program criteria, not narrowly optimized against a single endpoint.

For the field more broadly, this changes which programs are economically viable. Targets and modalities once considered too risky or too expensive — rare diseases, difficult target classes, peptides and biologics with narrow PK windows — become feasible when the number of experiments required to reach a candidate drops. The question stops being how many experiments a program can afford, and becomes how few it actually needs.

“We are entering an era where the cost of a drug discovery program will be measured not in the number of experiments, but in the quality of decisions behind them,” said Dr. Alan Nafiiev, Founder & CEO of Receptor.AI. “With StratAI™ at the core of PharmaSphere™, our partners reach validated candidates in fewer cycles and with full traceability of how every choice was made — and that fundamentally changes which programs are worth starting in the first place.”