Streamlining Antibody Discovery: NOS (Guided Diffusion for CDR Design) is Now Available on Vecura

What is NOS?

NOS (diffusioN Optimized Sampling) is a sophisticated sequence-space diffusion framework designed for the precise engineering of antibody complementarity-determining regions (CDRs). By operating directly in sequence space, the model utilizes a denoising transformer architecture to generate, redesign, and optimize antibody sequences starting from VH/VL seed pairs. It uniquely bridges generative modeling with biophysical constraints, allowing researchers to explore sequence diversity or steer evolution toward specific stability and structural targets.

It helps users perform targeted antibody design without the need for extensive manual sequence manipulation. It is especially useful for antibody discovery teams focusing on affinity maturation and the optimization of biophysical properties such as solubility or structural content in therapeutic candidates.

What can users do with NOS on Vecura?

With NOS on Vecura, users can:

• Redesign CDRs: Selectively mask and replace CDR regions in heavy and light chain sequences to generate diverse libraries for experimental testing.
• Apply Gradient Guidance: Utilize biophysical objectives—such as solvent-accessible surface area (SASA) or secondary structure metrics—to steer the diffusion process toward optimized variants.
• Streamline Workflows: Leverage pre-configured inference pipelines (using standard numbering schemes like Chothia or AHo) to generate variants without managing local infrastructure.
• Compare Variants: Access detailed output files containing redesigned sequences and per-position masks to systematically filter and prioritize candidates for synthesis.

What the output means

The output provides a structured CSV table containing generated antibody variants, original seed sequences, redesign masks, and guidance hyperparameter settings. This output should be used to support scientific decision making, helping researchers narrow down vast sequence spaces to the most promising candidates. It does not replace experimental validation.

Why this matters

Antibody design traditionally relies on high-throughput screening or heavy computational simulation, which can be both time-consuming and resource-intensive. NOS addresses this by incorporating discriminative gradient guidance directly into the generative process, allowing for "design-by-objective" at the sequence level. This accelerates the path from an initial lead antibody to a refined, optimized molecule, potentially reducing the number of wet-lab iterations required to achieve target binding and expression levels.

By integrating this model into a cloud platform, users gain immediate access to complex diffusion-based sampling without the technical overhead of training custom generative models or managing complex dependency chains.

• Developed by: Research group associated with the LaMBO-2 study.
• Source: Official GitHub repository
• Reference: Gruver et al., arXiv:2305.20009