Abstract
Drug discovery has long sought computational systems capable of designing drug-like molecules directly: developable and non-immunogenic from the start. Here we introduce Latent-X2, a frontier generative model that achieves this goal through zero-shot design of antibodies with strong binding affinities, drug-like properties, and, for the first time for any de novo generated antibody, confirmed low immunogenicity in human donor panels. Latent-X2 is an all-atom model conditioned on target structure, epitope specification, and optional antibody framework, jointly generating sequences and structures while modelling the bound complex. Testing only 4 to 24 designs per target in each modality, we successfully generated VHH and scFv antibodies against 9 of 18 evaluated targets, achieving a 50% target-level success rate with picomolar to nanomolar binding affinities. Designed molecules exhibit developability profiles that match or exceed those of approved antibody therapeutics, including expression yield, aggregation propensity, polyreactivity, hydrophobicity, and thermal stability, without optimization, filtering, or selection. In the first immunogenicity assessment of any AI-generated antibody, representative de novo VHH binders targeting TNFL9 exhibit both potent target engagement and low immunogenicity across T-cell proliferation and cytokine release assays. The model generalizes beyond antibodies: against K-Ras, long considered undruggable, we generated macrocyclic peptide binders competitive with trillion-scale mRNA display screens. These properties emerge directly from the model, demonstrating the therapeutic viability of zero-shot molecular design, now available without AI infrastructure or coding expertise at https://platform.latentlabs.com.
Introduction
In traditional drug discovery, finding a molecule that binds to a disease target is only half the battle. Many candidates fail later in development not because they lack binding affinity, but because they have poor "developability" (e.g., they aggregate, are unstable, or are hard to manufacture) or high immunogenicity (they trigger an unwanted immune response in the patient). Fixing these liabilities usually requires lengthy, expensive optimization cycles that often result in zero-sum trade-offs—improving stability might ruin binding affinity, for example. The holy grail of computational drug design is a system capable of "zero-shot" generation: designing molecules that bind tightly, behave well biophysically, and evade the immune system right out of the gate, without needing post-generation optimization.
Methods
Latent-X2 is an all-atom generative model that jointly designs the 3D structure and amino acid sequence of a binder. The model is conditioned on the target protein's structure, specific "hotspot" residues (the desired binding epitope), and an optional framework (like a specific VHH or scFv antibody scaffold). Unlike previous models that require task-specific fine-tuning, Latent-X2 uses a single architecture to generate diverse modalities, including single-domain antibodies (VHHs), single-chain variable fragments (scFvs), and macrocyclic peptides. Generated candidates are filtered using structure prediction models (like AlphaFold or Chai-1) and confidence metrics (like DockQ) to select the most viable designs for synthesis.
Key Results
The model was tested against 18 diverse soluble protein targets for antibodies and achieved a 50% target-level success rate, testing only 4 to 24 designs per target. Affinities were exceptionally strong, with the best binder achieving a $K_D$ of 26.2 pM against HDAC8.
Beyond antibodies, Latent-X2 demonstrated significant capability in designing macrocyclic peptides. For the target PHD2, 9 out of 10 tested designs bound successfully, with the best achieving 1.54 nM affinity, significantly outperforming the 729 nM affinity of the best reported hit from trillion-scale mRNA display screens.
Furthermore, 80% of confirmed binders met strict developability thresholds derived from approved therapeutics. In a major milestone, four VHH binders targeting TNFL9 showed zero detectable immunogenic response in human T-cell proliferation or cytokine release assays, marking the first time an AI-generated antibody has been confirmed to have low immunogenicity in human donor panels.
Discussion
The emergent developability of Latent-X2 designs suggests the model has deeply learned the underlying atomistic biochemistry of stable, non-promiscuous binding. By achieving competitive results with just 10 designs compared to trillion-scale experimental screens, Latent-X2 represents an 11-order-of-magnitude reduction in experimental search space. While ex vivo human donor panels are a strong proxy, future work will involve in vivo studies and human clinical trials to further validate these therapeutic leads.
Summary updated