The scFv-versus-nanobody choice usually gets made by habit. Someone reaches for whichever fragment their lab already knows, or picks the one that sounds more like a “real” antibody. That is the wrong axis. The decision that matters is not which molecule is more sophisticated, it is where the molecule has to end up. Get the endgame right and the fragment, and the tool that designs it, follow almost automatically.
This post lays out the real tradeoff, then maps each fragment to the AI tool that designs it.
The endgame decides the fragment
A nanobody, or VHH, is a single antibody domain of roughly 15 kDa. It is the variable region of a heavy-chain-only antibody, with no light chain to pair. That makes it small, stable, easy to express in yeast or bacteria, and quick to screen by display. Its long CDR-H3 loop can reach into concave or cryptic sites, enzyme active sites, receptor clefts, that a flatter paratope cannot.
An scFv is two domains, a heavy variable region and a light variable region, joined by a linker into roughly 25 kDa. It carries all six CDRs, so it presents a larger, more conventional antibody paratope. Critically, it is built on a paired-chain human framework, which means it reformats cleanly into a full IgG. That is the property you are paying the extra complexity for.
So the two fragments are not ranked, one is not the grown-up version of the other. They are two different starting points for two different destinations.
How we actually choose
The mistake is treating the format as a matter of taste. In practice it is a matter of where the program is headed.
Reformatting and humanization are both expensive, so you want to start from the fragment already close to the destination. If the destination is a human IgG, an scFv on a human framework is most of the way there. If the destination is a small, fast, multivalent binder, a nanobody is already the product.
Which tool designs which
Here is the practitioner payoff, because the fragment you pick determines the tool you run. On the Ranomics platform these are two distinct tools, one per format.
VHH, use RFantibody. RFantibody is an antibody-finetuned RFdiffusion model. It holds a nanobody framework fixed and diffuses the heavy-chain CDR loops (H1, H2, H3) against your target and hotspots, recovers sequences with ProteinMPNN, and filters each design by re-predicting the complex with an antibody-tuned RoseTTAFold2. The output is a real single-domain VHH, ready for the humanization and multivalent-format work that follows.
scFv, use ESMFold2 design. This tool inverts the ESMFold2 structure prediction model: it runs gradient descent on a soft sequence representation, backpropagated through the fold network, to design all six CDRs at once on a locked humanized framework. You pick the framework, trastuzumab, atezolizumab, or ocankitug, all clinically validated humanized antibodies, and the gradient descent only mutates the CDRs. It is the only tool in the catalog that designs paired heavy-and-light scFv CDRs end to end, and it was wet-lab validated against PDGFRB, EGFR, PD-L1, CD45, and CTLA4 with nanomolar affinity.
Notice how the tool choice reinforces the format logic. The scFv tool starts from a clinically validated human framework, so the design is humanization-ready by construction, which is exactly what the IgG endgame needs. The VHH tool gives you a fast single domain, which is exactly what the efficient-default case wants.
The trap: deciding the format late, or by fashion
The expensive mistake is leaving the format decision until after design, or making it on vibes. Two ways it bites.
Design a nanobody, then discover the program needs a bivalent human IgG, and you have a reformatting and humanization job standing between your validated binder and the actual product. Or default to an scFv because it looks more like a therapeutic antibody, when a nanobody would have been faster to design and screen and the endgame never required a paired chain, and you have paid for complexity you did not use. Decide the endgame first, then the fragment falls out, then the tool.
And the trap that applies to either fragment: a designed CDR set is a candidate, not a binder. Both RFantibody and ESMFold2 design filter on in silico confidence, a re-predicted complex or a converged fold loss, which ranks plausibility, not binding. Whether the fragment actually engages its target is a question only a display screen answers.
Quick reference
| Nanobody (VHH) | scFv | |
|---|---|---|
| Size and chains | ~15 kDa, single domain, no pairing | ~25 kDa, paired VH + VL |
| Design tool | RFantibody (CDR loops on a fixed framework) | ESMFold2 design (six CDRs via inversion) |
| Paratope reach | long CDR-H3, good for concave and cryptic sites | six CDRs, good for larger flat interfaces |
| Humanization | camelid framework, needs humanizing | designed on a human framework, humanization-ready |
| Best endgame | fast, small, multivalent or bispecific binders | a full human IgG or paired-chain antibody |
Practical recommendations
- Decide the endgame before the fragment. If the product is a human IgG, start with an scFv. If it is a fast, small, or multivalent binder, start with a VHH.
- Default to the nanobody when the endgame is open. It is cheaper to design, express, and screen, and it is simple to build up into larger formats later.
- Match the tool to the fragment. RFantibody designs the VHH; ESMFold2 design designs the scFv. They are different models for different formats, not competitors.
- Confirm binding at the bench, not in the score. A confident CDR design is a candidate. Take it to display, then to the bench.
Design a VHH with RFantibody or an scFv with ESMFold2 design on tools.ranomics.com, and if you are still choosing the fragment, the RFantibody nanobody guide covers the single-domain case in more depth.
Summary
scFv versus nanobody is a decision about the endgame, not about which fragment is more sophisticated. A VHH is the faster, cheaper default, designed with RFantibody, and it is the right call when the product is a small or multivalent binder. An scFv is built on a human framework and reformats cleanly to IgG, designed with ESMFold2 inversion, and it earns its extra cost when the endgame is a full human antibody. Decide where the molecule has to end up, and the fragment and the tool follow. Then remember that either way, no design is a binder until a screen says so.
Related Ranomics services
- RFantibody for de novo nanobody design: the VHH side, how CDR loops are diffused onto a fixed nanobody framework.
- Nanobody discovery by yeast display: the experimental step where a designed fragment gets measured, not just predicted.
- Binder Pilot: take a shortlist of designed VHH or scFv candidates to a ranked, wet-lab-validated hit list on a fixed-scope campaign.