Use OpenGerminal Online

OpenGerminal: Open-Source Pipeline for De NovOpenGerminal: Open-Source Pipeline for De Novo VHH Antibody Design

OpenGerminal is an open-source implementation of the Germinal antibody design pipeline that designs nanobody (VHH) binders against a target antigen using only open-source dependencies. It was developed by Bing Han and Sheng Li at the University of Virginia (Center for Membrane and Cell Physiology; Department of Molecular Physiology and Biological Physics; School of Data Science). The code is released under the Apache 2.0 license and is also distributed as an Apptainer container.

Germinal combines gradient-based sequence hallucination with structure cofolding and CDR redesign to generate antibody binders. According to the authors, the original pipeline depends on PyRosetta and the IgLM language model, whose licenses restrict use and redistribution. OpenGerminal swaps these out so the workflow can be used and shared more freely.

How OpenGerminal Works

The Four-Stage Design Pipeline

1. Sequence design: Gradient-based hallucination with ColabDesign's AlphaFold-Multimer optimizes CDR sequences against structural objectives (pLDDT, iPTM, PAE) and a language-model naturalness score.

2. Initial cofolding filter: Chai-1 re-predicts each design, followed by structural relaxation and interface scoring. Geometric failures are discarded.

3. CDR redesign: AbMPNN diversifies the CDR sequences on the hallucinated backbone; each seed can yield up to four sequence variants.

4. Final filter: Chai-1 re-folds the redesigns, and stricter structural and physicochemical thresholds are applied.

What Replaces PyRosetta

• Relaxation: OpenMM 8.5.1 with FASPR side-chain repacking.

• Interface analysis: FreeSASA for surface area, Biopython for geometric analysis, and sc-rs v1.0.0 for shape complementarity.

• Compatibility: All nine interface functions of the original PyRosetta utility module are provided through an API-compatible backend.

Benchmark Results Against Germinal

The authors compared the two pipelines on PD-L1 and IL-3 using the same hotspots and VHH framework. Per-trajectory hallucination was slower in OpenGerminal (4.4 vs. 3.0 minutes on PD-L1 and 4.2 vs. 2.6 minutes on IL-3, on an NVIDIA A100 80 GB).

Pipeline Yield

No accepted designs were obtained for IL-3 with either pipeline.

Quality of Accepted PD-L1 Designs

OpenGerminal produced 18 accepted structures from 9 seeds, versus 16 structures from 7 seeds for Germinal. Median Chai-1 metrics were slightly higher for OpenGerminal.

Neither set showed steric clashes, and CDR3-hotspot contact counts were similar. In a post-hoc check against PyRosetta, the open-source shape complementarity score correlated at Spearman rho 0.882.

What is Tamarind Bio?

Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.

The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:

• A user-friendly graphical interface for setting up and launching experiments

• A robust API for integration into existing research pipelines

• An automated system for managing and scaling computational resources

Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.

Accelerating Discovery with OpenGerminal on Tamarind Bio

• Nanobody design for a defined epitope: Specify hotspot residues on an antigen and generate candidate VHH binders.

• Open and redistributable workflows: Use a design pipeline free of the PyRosetta and IgLM licensing constraints described by the authors.

• Comparing against Germinal: The authors benchmarked both pipelines on PD-L1 and IL-3 with matched inputs.

• Candidate generation for follow-up: Accepted designs are returned as PDB files with scoring metrics for further triage.

How to Use OpenGerminal on Tamarind Bio

1. Log in and open the tool: Sign in at tamarind.bio and select OpenGerminal.

2. Provide the target antigen: Supply the antigen structure; single-chain targets are the validated case.

3. Specify hotspot residues: Indicate the epitope residues the binder should engage.

4. Choose a VHH framework: Select the nanobody framework to be designed.

5. Adjust settings and run: Set any available options, such as the number of trajectories, then submit the job.

6. Download outputs: Retrieve accepted designs as PDB files and the metrics CSV.

7. Interpret results: Review Chai-1 confidence metrics such as pLDDT and iPTM, and treat placeholder metrics with caution.

8. Use downstream: Prioritize designs for further in silico analysis or experimental testing.

Things to Keep in Mind

• Hallucination is about 1.5-1.6 times slower per trajectory than Germinal, and more designs reach the costly later stages.

• binder_score (fixed at -1.0) and interface_dG (fixed at -10.0) are hard-coded placeholders; Stage 2 therefore always selects the first structure.

• interface_hbonds is fixed at 5, so the Stage 4 hydrogen-bond filter always passes.

• Multi-chain support is not fully validated; an insulin test yielded no accepted designs.

• Only two targets were benchmarked, and the comparison with Germinal is confounded by different relaxation methods.

• No experimental validation of OpenGerminal designs has been reported.

Source: Han B, Li S. OpenGerminal: an open-source implementation of the Germinal antibody design pipeline. bioRxiv 2026. DOI: 10.64898/2026.06.25.734527. Code: https://github.com/teaninja/OpenGerminal.o VHH Antibody Design