Harnessing Vecura’s AI Platform to Design FGFR4-Targeting Peptides for Hepatocellular Carcinoma Therapy

Introduction

Hepatocellular carcinoma (HCC) is the predominant primary liver cancer and a major cause of cancer-related mortality worldwide [1]. Despite advances in current therapies, advanced HCC remains difficult to treat, highlighting the need for new therapeutic targets. The FGF19–FGFR4 signaling axis plays an important role in HCC progression by activating the RAS–RAF–MEK–ERK and PI3K–AKT pathways, thereby promoting cancer-cell proliferation, survival, migration, and invasion [2], [3], [4]. Therefore, we propose to design a peptide that specifically binds the extracellular domain of FGFR4 near the FGF19 recognition interface to block FGF19-dependent signaling and suppress HCC progression (Figure 1).

Figure 1. FGF19–FGFR4 Signaling and Proposed Peptide Blockade

To translate this therapeutic concept into experimentally testable candidates, a structure-guided peptide design strategy can be implemented using the Vecura platform. By integrating protein structure prediction, peptide generation, molecular interaction analysis, and in silico evaluation within a unified workflow, Vecura enables the systematic design and prioritization of FGFR4-targeting peptides before experimental testing. Candidate peptides can be designed against the extracellular surface of FGFR4 surrounding the FGF19 recognition interface and subsequently evaluated for predicted binding affinity, structural stability, and target specificity. This computational workflow provides a practical bridge from molecular design to experimental validation, helping prioritize a focused set of candidates for subsequent biochemical and cellular assays.

Table 1. FGFR4 Peptide Binder Discovery Pipeline

Workflow

Try the workflow here

Design and Candidate Selection

De novo peptide binders were generated against the Ig-like D3 domain of human FGFR4 (UniProt P22455) using BoltzProt in generate_protein_binder mode. The target construct contained 97 residues, including twenty pre-defined epitope positions concentrated in the βC–βC′ loop (P279–Q303) and the LAGNSIG motif (Y331–S342). Fifty designs ranging from 20 to 35 residues were produced without template guidance, and cysteine was excluded. Across the full set, interface confidence ranged from 0.000 to 0.651 (median, 0.450), structure confidence ranged from 0.464 to 0.727 (median, 0.587), and ipTM ranged from 0.839 to 0.976 (median, 0.960). Interface confidence correlated positively with ipTM (r = +0.73) and negatively with minimum interaction PAE (r = −0.66). Only four of the fifty designs simultaneously met the thresholds for binding confidence (≥ 0.50) and structure confidence (≥ 0.65). A composite ranking score based on four orthogonal metrics was therefore used to select the five best designs for further evaluation.

Table 2. Top Five Candidates After Scoring All 50 Candidates Across Four Metrics

Structural and Energetic Validation

Complex coordinates were extracted for the five top-ranked designs, and epitope coverage was measured directly from the 3D structures using a 5.0 Å heavy-atom contact cutoff rather than inferred from sequence. Two candidates—Top 2 (PTEEQIEEFYRKLFEELFGITIK) and Top 3 (PPQMSTEEMVQEYQKLVKEKLAE), both 23 residues long—each engaged 10 of the 20 target epitope residues and were selected for independent refolding with Boltz-2.1 in triplicate. Both reproduced their predicted poses with high consistency (Top 2: ipTM, 0.910 ± 0.001; structure confidence, 0.917 ± 0.001; Top 3: ipTM, 0.866 ± 0.012; structure confidence, 0.883 ± 0.015), supporting the stability of the predicted binding modes across independent sampling. PRODIGY estimated low-micromolar predicted affinity for both candidates, with Top 3 showing marginally stronger predicted binding (ΔG, −8.08 kcal/mol; Kd, 1.99 µM; 61 interfacial contacts) than Top 2 (ΔG, −7.71 kcal/mol; Kd, 3.68 µM; 64 interfacial contacts).

Table 3. Independent Refolding Data for Top 2 and Top 3

Developability Profiling and Outcome

Both candidates passed the computational safety and formulation filters. ToxinPred3, using the hybrid ML/MERCI mode, classified both sequences as non-toxic, with scores well below the 0.38 threshold (Top 2, 0.185; Top 3, 0.000). NetSolP-1.0 also predicted good aqueous solubility for both candidates (Top 2, 0.892 ± 0.042; Top 3, 0.808 ± 0.067). The main differences between the candidates concerned their liability profiles rather than their predicted potency. Top 2 contains no methionine and therefore has no predicted methionine-oxidation liability, but it showed a higher amyloid-aggregation score (PALM, 0.623; hotspot at residues 15–23) and a higher instability index (83.7). Top 3 showed a more favorable aggregation and stability profile (PALM, 0.494; instability index, 67.7), but contains two oxidation-prone methionines at positions 4 and 9. Because neither candidate clearly outperformed the other across the twelve assessed criteria (6:5 in favor of Top 3), both are recommended for parallel chemical synthesis and experimental testing. SPR or BLI against FGFR4 and the FGFR1/2/3 paralogs should serve as the decisive go/no-go step.

Table 4. Liability Profiles and Physicochemical Properties

How Vecura Helps

Vecura helps researchers move from a therapeutic hypothesis to experimentally testable peptide candidates through an integrated, structure-guided design workflow. For the FGF19–FGFR4 axis, the platform supports the identification of relevant protein structures and interaction interfaces, the generation of FGFR4-targeting peptide candidates, and the computational evaluation of predicted binding affinity, structural stability, and specificity. By bringing these design and assessment steps into a unified workflow, Vecura helps researchers prioritize promising candidates, identify potential liabilities earlier, and reduce unnecessary experimental screening. Ultimately, Vecura supports a more systematic transition from molecular design to biochemical and cellular validation, accelerating the development of peptide-based strategies to block FGF19-driven signaling.

From Computational Design to Experimental Validation

This study demonstrates how Vecura can support a systematic, structure-guided approach to designing FGFR4-targeting peptides for the potential inhibition of FGF19-driven signaling in HCC. By combining de novo peptide generation, structural validation, binding-affinity prediction, and developability profiling, the workflow enabled the prioritization of two promising candidates, Top 2 and Top 3, for further investigation. Both candidates showed consistent predicted binding poses, favorable predicted affinity, and acceptable computational safety and solubility profiles, although each presented distinct developability considerations. These findings highlight the value of integrating multiple computational criteria when selecting peptide candidates for experimental testing. The next step is to validate FGFR4 binding and receptor specificity experimentally, followed by assessment of their ability to inhibit FGF19-dependent signaling and tumor-associated cellular phenotypes. Ultimately, Vecura provides a practical bridge between therapeutic hypothesis and experimental validation, helping researchers move from computational peptide design toward more focused and informed development of peptide-based therapeutics.

References

1. Hwang, S. Y.; Danpanichkul, P.; Agopian, V.; Mehta, N.; Parikh, N. D.; Abou-Alfa, G. K.; Singal, A. G.; Yang, J. D., Hepatocellular carcinoma: updates on epidemiology, surveillance, diagnosis and treatment. Clinical and Molecular Hepatology 2025, 31 (Suppl), S228-S254.

2. Kuzina, E. S.; Ung, P. M.-U.; Mohanty, J.; Tome, F.; Choi, J.; Pardon, E.; Steyaert, J.; Lax, I.; Schlessinger, A.; Schlessinger, J.; Lee, S., Structures of ligand-occupied β-Klotho complexes reveal a molecular mechanism underlying endocrine FGF specificity and activity. Proceedings of the National Academy of Sciences 2019, 116 (16), 7819-7824.

3. Liu, Y.; Cao, M.; Cai, Y.; Li, X.; Zhao, C.; Cui, R., Dissecting the Role of the FGF19-FGFR4 Signaling Pathway in Cancer Development and Progression. Frontiers in Cell and Developmental Biology 2020, 8.

4. Zhan, T.-A.; Xia, F.; Huang, H.-W.; Zhan, J.-C.; Liu, X.-K.; Cheng, Q., Fibroblast growth factor 19-fibroblast growth factor receptor 4 axis: From oncogenesis to targeted-immunotherapy in advanced hepatocellular carcinoma. World Journal of Gastrointestinal Oncology 2025, 17 (9).