Latent-Y

Latent-Y is the first lab-validated autonomous AI agent for drug design, developed by Latent Labs. Powered by Latent-X2, Latent Labs' frontier model for drug-like antibody design, Latent-Y enables researchers to design novel antibodies from plain text prompts — compressing weeks of expert computational work into hours. It is designed for drug discovery teams seeking to dramatically expand the scale and speed at which they can explore therapeutic opportunities, without requiring deep computational expertise.

Latent-Y operates in the same environment as protein design experts, with access to bioinformatics tools, biological databases, and external scientific publications. It applies expert-level reasoning to navigate autonomously from a research objective to lab-ready candidates. In user studies, experts working with Latent-Y completed design campaigns 56-fold faster than independent expert time estimates. Across nine targets, the agent produced lab-confirmed binders against six, achieving a 67% target-level success rate with binding affinities reaching the single-digit nanomolar range — all without human filtering or intervention at any stage.

Validated Campaign Types

• Epitope discovery: Given a therapeutic specification — such as a desired mechanism of action, a functional outcome, or constraints on the binding site — Latent-Y applies biological reasoning to identify epitopes matched to those goals and proceeds to design accordingly.
• Cross-species binder design: Latent-Y generates antibodies that bind homologous targets across species, supporting the translational studies required to progress a program toward the clinic. In demonstrated campaigns, the agent designed nanobodies that simultaneously bound human and cynomolgus homologs, confirmed in the laboratory.
• Design from publication: Given a scientific paper as input, Latent-Y identifies relevant targets and epitopes, reasons about the published mechanism of action, and autonomously designs binders. In one demonstrated campaign, the agent processed a publication on blood-brain barrier crossing and designed antibodies targeting human transferrin receptor (hTFR1), confirmed in the laboratory without any manual curation of the input.

Key Features and Capabilities

• Flexible autonomy with full transparency: Latent-Y can run fully autonomously end-to-end, or pause at each stage to surface progress summaries and recommended next steps for scientist review. Every design decision is recorded as interpretable reasoning that scientists can evaluate, challenge, and build on.
• Lab-validated end-to-end results: Lab-confirmed binders across six of nine targets attempted, with affinities reaching the low nanomolar range. Every confirmed binder is a novel molecule designed de novo from a text prompt.
• Force multiplier for experts: User studies demonstrate a 56-fold expert acceleration, compressing weeks of expert computational work into hours of autonomous operation.
• Parallel campaign execution: A single researcher can run multiple design campaigns simultaneously across targets and modalities, transforming the scale at which a drug discovery team can explore therapeutic opportunities.
• Natural language and publication inputs: The agent can process natural language goals, research work plans, or scientific publications directly as campaign inputs, requiring no coding or manual curation.
• Powered by Latent-X2: The underlying generative model produces drug-like antibodies with drug-like developability, enabling difficult targets and complex design goals from the first generation.
• Broad modality coverage: The same agent architecture supports VHH, macrocyclic peptide, and mini-binder design campaigns.

Workflow Overview

1. A researcher submits a text prompt specifying design goals, constraints, or a scientific publication as input.
2. Latent-Y autonomously reasons about the biological context, identifies relevant targets and epitopes, and formulates a design strategy.
3. The agent executes the full design campaign, leveraging bioinformatics tools, biological databases, and Latent-X2 to generate novel antibody candidates.
4. Computationally passing binder candidates are surfaced as lab-ready outputs, with all design decisions logged and available for scientist review.
5. Scientists can optionally engage in human-in-the-loop collaboration at any stage, reviewing progress summaries and recommended next steps.

Latent-Y is available to selected partners through the Latent Labs Platform, offering a complete lab-validated, no-coding workflow. All results are preclinical; Latent-Y accelerates the computational stages of drug discovery and does not replace the experimental stages that must follow. Future development directions include closing the loop with experimental feedback, expanding the agent's action space, and integrating with robotic laboratories. Commercial and partnership enquiries can be directed to [email protected].