molence bio
Platform
Home / Research / Mol-1

Design meets experiment

From a predicted interaction to a molecule that binds, activates, and improves with feedback.

A molecular design becomes exciting when it starts doing something in the lab. Can it recognize the intended target? Trigger a useful response? Help us improve our model for the next round of design? We built a VHH design workflow around an early version of Mol-1 to follow that journey—from computational ideas to molecules with measurable activity.

VHHs are naturally single-domain antibodies, making them well suited as modular building blocks for more complex therapeutics. Across our campaigns, they gave us a way to explore several ambitions at once—binding a chosen site, activating a receptor, recognizing two antigens, and learning from experiments.

Finding the target—and the right place to bind

Our first milestone was binding across a wide range of targets. Designed VHHs produced binding hits against 17 targets, ranging from soluble proteins and receptor extracellular domains to GPCRs. GPCRs are important therapeutic targets that have long been challenging for conventional antibody discovery. Initial hit rates reached 63% (Figure 1). A single design workflow was opening doors across very different molecular landscapes.

Designed VHH–target complex models: receptor extracellular domains, GPCRs, and benchmark targets

Where a molecule binds can matter as much as whether it binds. For TNFR2 and IL-4Rα, we aimed at the regions used by their natural ligands. The designed VHHs competed with those ligands for binding (Figure 1). These results support binding to the intended ligand-binding regions, highlighting the potential of de novo VHH design to target chosen epitopes. By building that specificity into the design process, we could streamline therapeutic binder discovery and reduce the need for downstream epitope screening.

Binding hit rates
020406070%PDGFR: 63%PDGFR63%14-3-3ε: 53%14-3-3ε53%IL-20: 50%IL-2050%CCR8: 44%CCR844%UBC9: 43%UBC943%BCMA: 40%BCMA40%Cytokine A: 37%Cytokine A37%CXCR4: 32%CXCR432%SAE1: 23%SAE123%Cytokine B: 23%Cytokine B23%PD-L1: 20%PD-L120%SORCN: 20%SORCN20%IL-4Rα: 20%IL-4Rα20%TNFR2: 10%TNFR210%4-1BB: 7%4-1BB7%CXCR6: 6%CXCR66%CCR5: 6%CCR56%
TNFR2
TNFR2 designed VHH and natural ligand structural overlay TNFR2 sequential ligand competition BLI 0 500 1000 Time (s) 0 2 4 6 8 Response (nm) VHH Buffer TNF- α
IL-4Rα
IL-4Rα sequential ligand competition BLI 0 500 1000 Time (s) 0 2 4 6 8 Response (nm) VHH Buffer IL-4
Figure 1. Recognition across a diverse target landscape. Designed complexes, binding-hit rates across 17 targets, and competition with natural ligands at TNFR2 and IL-4Rα.

Building molecules worth taking forward

An exciting binding result becomes more valuable when the molecule also has favorable physical properties. We looked across hydrophobicity, thermal stability, self-interaction, and nonspecific binding to see how often those qualities came together.

Across 30 fully profiled candidates, 23 met at least three of the four criteria. Those candidates covered 12 of 13 targets (Figure 2). Favorable properties appeared across a broad set of designs, giving us more starting points to carry into the next stage of molecular engineering.

Four-assay screening of 30 VHH-Fc candidates across 13 targets Target-level ≥3 criteria Candidate-level ≥3 criteria 0 25 50 75 100 Proportion (%) 92.3% (12/13) 76.7% (23/30) a Combined profile Pass rate: 80.0% 0 5 10 15 Retention time (min) b HIC Pass rate: 80.0% 40 50 60 70 80 Reported Tm1 (°C) c DSF Pass rate: 83.3% 0 5 10 15 Wavelength shift (nm) d AC-SINS Pass rate: 70.0% 0 25 50 75 100 Score e BVP
Figure 2. Biophysical screening across 13 targets. Thirty VHH-Fc candidates were assessed across four assays. Panel a shows that 23/30 candidates, spanning 12/13 targets, met at least three criteria. Panels b–e show hydrophobicity (HIC), thermal stability (DSF), self-interaction (AC-SINS), and nonspecific binding (BVP). Dots represent individual candidates; green shading and dashed lines mark screening ranges and thresholds. Pass rates are shown below each assay panel.

From high-affinity binding to cellular activity

4-1BB: compact molecules with potent agonist activity

4-1BB (CD137) helps T cells activate, multiply, and survive, making it an attractive target for cancer immunotherapy (Singh et al., 2024). Tumor-targeted and bispecific 4-1BB agonists offer a way to bring this immune-stimulating activity into multifunctional therapeutics (Klein et al., 2024).

For our 4-1BB designs, we wanted binding to translate into receptor activation. Two VHHs did exactly that: both showed greater agonist potency than Utomilumab in the reporter readout, with activation EC50 values of 0.05 and 0.22 nM, compared with 0.43 nM for Utomilumab (Figure 3).

The appeal extends beyond potency. These designs also combined high purity, thermal stability, and favorable molecular properties. Together, those features make them attractive starting points for building compact 4-1BB modules into bispecific and multispecific antibodies.

a · 4-1BB VHH 01 SPR
a · 4-1BB VHH 01 SPR 0 100 200 300 Time (s) 0 25 50 75 Response (RU) K D  = 3.5  nM 50 nM 25 nM 12 nM 6.2 nM 3.1 nM 1.6 nM
b · 4-1BB VHH 02 BLI
b · 4-1BB VHH 02 BLI 0 100 200 300 Time (s) 0.00 0.05 0.10 0.15 Response (nm) K D =103 nM 500 nM 250 nM 125 nM 62.5 nM 31.2 nM 15.6 nM
c · 4-1BB reporter activation
c · 4-1BB reporter activation 10 −4 10 −2 10 0 10 2 Antibody (nM) 0 4 8 12 RLU ( 10 6 ) VHH 01 VHH 02 Utomilumab Negative control
d · Activity and molecular profiles
CandidateActivation EC50
(nM)		SEC purity
(280 nm, %)		HIC RT
(min)		DSF Tm1
(°C)		AC-SINS
(nm)
4-1BB VHH 010.051007.4470.50
4-1BB VHH 020.22976.2370.21
Figure 3. Potent activation in a compact format. Two 4-1BB VHHs combine target binding, strong agonist activity, and favorable molecular profiles.

One round of feedback changed the next round of design

The first 4-1BB campaign produced two functional hits from 30 designs. We brought those experimental results back into the design process. In the next round, 51 of 60 designs were functional hits—a rise from 7% to 85% (Figure 4).

The next round delivered a much larger pool of potent agonists. This is the part of the story we find especially compelling. Experiments help shape the next set of molecules, turning individual successes into a broader starting point for further molecular engineering.

4-1BB two-round design–experiment loop 3 0.3 0.03 Concentration (nM) 0 2 4 6 8 Negative control Positive control Round 1 Round 2 Round 1 Round 2 0 20 40 60 80 100 Design success rate (%) 2/30 (7%) 51/60 (85%) b a Luminescence (×10 6  RLU)
Figure 4. From 7% to 85% functional hits. One round of experimental feedback expanded the pool of active 4-1BB designs and delivered strong activation in the next round.

BCMA: reaching 43 pM affinity from the first design round

BCMA is an established therapeutic target in multiple myeloma, with T-cell engagers and CAR-T therapies already putting it to work. Binding affinity is an important design parameter for BCMA-directed T-cell engagers (Chen et al., 2021). Cilta-cel uses two BCMA-binding VHH domains, illustrating the therapeutic value of compact, single-domain recognition modules (Banerjee et al., 2022).

BCMA offered another striking result. From 30 zero-shot designs, Mol-1 yielded 12 binding hits. One VHH reached an affinity of 43 pM; a second reached 1.1 nM (Figure 5). Here, a computational starting point led directly to exceptionally tight binding.

Both molecules also recognized BCMA on cells. Their cell-binding EC50 values were approximately 22- and 26-fold lower than the belantamab biosimilar comparator. Strong recognition in a compact format provides useful building blocks for multivalent and multispecific antibody designs.

a · BCMA VHH 01 SPR
a · BCMA VHH 01 SPR 0 100 200 300 Time (s) 0 20 40 Response (RU) K D  = 43  pM 6.2 nM 3.1 nM 1.6 nM 0.78 nM 0.39 nM
b · BCMA VHH 02 SPR
b · BCMA VHH 02 SPR 0 100 200 300 Time (s) 0 20 40 Response (RU) K D  = 1.1  nM 12 nM 6.2 nM 3.1 nM 1.6 nM 0.78 nM
c · BCMA cell binding
c · BCMA cell binding 10 −2 10 −1 10 0 10 1 Antibody (nM) 0 1 2 3 MFI ( 10 6 ) Cell­binding  EC 50 VHH 01: 0.19 nM VHH 02: 0.16 nM Belantamab: 4.1 nM
Figure 5. 43 pM affinity, with strong recognition on cells. Two BCMA VHHs combine tight target binding with lower cell-binding EC50 values than the belantamab biosimilar comparator.

CXCR6: taking design to a membrane receptor

We also took the workflow to CXCR6, a seven-transmembrane chemokine receptor. The designed VHH recognized CXCR6 on cells and antagonized signaling induced by its ligand, CXCL16 (Figure 6). A designed molecule had reached beyond target recognition to change a cellular response.

a · CXCR6 cell binding
a · CXCR6 cell binding 2 10 100 Antibody (nM) 0 5 10 15 MFI ( 10 3 ) EC 50  = 39  nM
b · CXCR6 antagonism
b · CXCR6 antagonism 50 75 100 125 Normalized cAMP response (%) 30 nM CXCL16 − + + − 167 nM VHH − − + +
Figure 6. From cell binding to functional antagonism. The CXCR6 VHH recognizes the receptor on cells and counteracts CXCL16-induced signaling.

One VHH, two antigens—by design

Recognizing two antigens with a single VHH is an appealing capability, but finding such molecules through conventional screening can be challenging. We made cross-reactivity a design objective from the outset, asking Mol-1 to create VHHs that recognize both cytokine-A and cytokine-B—two non-orthologous protein antigens sharing only 54% sequence identity.

Five of just 30 designs bound both antigens (Figure 7). To our knowledge, this is the first experimental demonstration of de novo-designed VHHs that cross-react with two non-orthologous protein antigens as intended.

These results show that dual-antigen recognition can be built into the design goal, offering a route to discovering cross-reactive binders with less downstream screening.

Cross-antigen affinities and VHH 01 six-concentration BLI 01 02 03 04 05 Cross­binding VHH 0 100 200 300 Apparent  K D, 1  (nM) 15 14 100 251 19 26 95 9 12 2 a Cytokine­A Cytokine­B 0 100 200 300 Time (s) 0.00 0.25 0.50 0.75 1.00 Response (nm) K D, 1  = 15 nM K D, 2  = 264 nM b VHH 01 · Cytokine­A Concentration (nM) 500 250 125 62.5 31.25 15.625 0 100 200 300 Time (s) 0.00 0.25 0.50 0.75 1.00 Response (nm) K D, 1  = 26 nM K D, 2  = 482 nM c VHH 01 · Cytokine­B Concentration (nM) 500 250 125 62.5 31.25 15.625
Figure 7. Dual-antigen recognition by design. Five VHHs bind both cytokine-A and cytokine-B; the binding curves show one design recognizing each antigen.

Building the next generation of medicines

These results show how biomolecular intelligence can translate into molecules with useful properties—from precise recognition and potent binding to cellular activity. They are an early expression of our broader ambition: to make AI a driving force in the discovery and development of new medicines.

As an AI-native biotech, we bring frontier AI, molecular engineering, and experimental science together to pursue targets and therapeutic possibilities that have remained out of reach. With Mol-1, we aim to push the frontiers of biomolecular intelligence for the benefit of all, bringing us closer to a future where every disease can be cured.

References

  • Singh, R., Kim, Y.-H., Lee, S.-J., Eom, H.-S., & Choi, B. K. (2024). 4-1BB immunotherapy: advances and hurdles. Experimental & Molecular Medicine, 56, 32–39. https://doi.org/10.1038/s12276-023-01136-4
  • Klein, C., Brinkmann, U., Reichert, J. M., & Kontermann, R. E. (2024). The present and future of bispecific antibodies for cancer therapy. Nature Reviews Drug Discovery, 23, 301–319. https://doi.org/10.1038/s41573-024-00896-6
  • Chen, W., Yang, F., Wang, C., Narula, J., Pascua, E., Ni, I., Ding, S., Deng, X., Chu, M. L.-H., Pham, A., Jiang, X., Lindquist, K. C., Doonan, P. J., Van Blarcom, T., Yeung, Y. A., & Chaparro-Riggers, J. (2021). One size does not fit all: navigating the multi-dimensional space to optimize T-cell engaging protein therapeutics. mAbs, 13, 1871171. https://doi.org/10.1080/19420862.2020.1871171
  • Banerjee, R., Lee, S. S., & Cowan, A. J. (2022). Innovation in BCMA CAR-T therapy: Building beyond the Model T. Frontiers in Oncology, 12, 1070353. https://doi.org/10.3389/fonc.2022.1070353