Individual Final Project: Computational Design of De Novo Anti-GFP Nanobodies Using the Germinal Framework

Presentation Link

Project HTGAA after Nanobdy Synthesis code

Abstract

This project applies the Germinal pipeline — an open-source computational framework combining AlphaFold-Multimer and the IgLM antibody language model — to design de novo nanobodies targeting a defined surface epitope of GFP. Four nanobody sequences were generated, codon-optimised via Asimov Kernel, annotated in Benchling, and ordered from Twist Biosciences as clonal genes for cell-free expression screening on the Ginkgo Bioworks Nebula platform.

Section 1 — Project Aims

Aim 1 — Experimental (HTGAA scope)

Design, synthesise, and confirm cell-free expression of de novo anti-GFP nanobodies using the Germinal pipeline and Ginkgo Bioworks Nebula.

Aim 2 — Development (PhD level)

Apply the validated pipeline to a clinically relevant target (NKG2D or PD-L1) and validate binding affinity by BLI.

Aim 3 — Visionary

Establish a fully democratised, open-source nanobody discovery platform accessible to any molecular biology laboratory without immunisation infrastructure.

Section 2 — Background

Note

**Source paper:** Mille-Fragoso et al. 2025 — *Efficient generation of epitope-targeted de novo antibodies with Germinal.* bioRxiv PMC12485712

Germinal achieves experimental success rates of 4–22% testing only 43–101 designs per target — dramatically fewer than phage display or random CDR mutagenesis.

Section 3 — Computational Design

Antigen preparation

GFP structure downloaded from PDB (2Y0G), cleaned with Biopython to remove solvent and chromophore atoms. Target epitope: residues A169–A171 (surface-exposed, confirmed by SASA analysis).

Framework

VHH framework from PDB: 1MEL (cAb-Lys3, anti-lysozyme). Chain A extracted, IMGT-numbered, CDR3 length verified at 14 residues.

Germinal run

Executed via Tamarind.bio hosted Germinal implementation. 20 designs generated. 4 passing designs selected by composite score (i_ptm, pLDDT, interface H-bonds, zero clashes).

Section 4 — Results

Top 4 designs

Rank	Design	i_ptm	pLDDT	H-bonds
1	s830310	0.798	0.890	12
2	s361430	0.784	0.919	8
3	s86205	0.770	0.895	11
4	s134340	0.751	0.897	6

DNA construct architecture

[T7 RBS 57bp] → [ATG] → [VHH 396bp] → [6xHis 15bp] → 
[Stop TAA] → [T7 terminator 67bp]
Total: 538 bp per construct

Benchling sequences

Google Collab Link

Section 5 — Methods

Tools used

Tip

All computational tools used in this project are open-source or HTGAA-partnered.

Tool	Purpose
google collab	All coding and data visualisation
Germinal (Tamarind.bio)	De novo CDR design
Biopython	PDB cleaning + SASA analysis
Asimov Kernel	Codon optimisation
Benchling	Sequence annotation + storage
SecureDNA	Biosafety screening
Twist Biosciences	Clonal gene synthesis
Ginkgo Nebula	Cell-free expression screening
py3Dmol	Structure visualisation

Expression construct

Vector:    pTwist Amp High Copy
System:    E. coli cell-free (PURExpress)
Detection: Anti-His HRP Western blot
Purification: Ni-NTA magnetic beads
Binding:   GFP pulldown assay

Section 6 — Ethical Considerations

Warning

All synthesised sequences are planned to be screened through SecureDNA prior to ordering. No biosecurity concerns identified — GFP is a non-pathogenic model antigen.

This project applies the principles of beneficence (reducing animal use in antibody generation) and justice (democratising access to molecular tools). All work will be conducted at BSL-1. No human subjects involved.

References

Mille-Fragoso et al. (2025) bioRxiv PMC12485712
Bennett et al. (2024) Science 385(6707)
Liu et al. (2020) NEJM 382:545
Evans et al. (2022) AlphaFold-Multimer bioRxiv
Shuai et al. (2023) Cell Syst 14:979