Clothianidin Binder Design with AffiniBind: A Neonicotinoid Target Campaign

GenX (HFPO-DA) Binder Design with AffiniBind

A ligand-conditioned binder campaign against GenX (HFPO-DA) generated 170 candidates, with a top docking score of -7.84 kcal/mol and a target QED of 0.806.

The Target Ligand

Clothianidin is a neonicotinoid insecticide with a chlorothiazole ring linked through a methylene bridge to a nitroimine-bearing guanidine group. Its SMILES string, CNC(=N[N+](=O)[O-])NCc1cnc(Cl)s1, encodes a compact polar molecule with a molecular weight of 249.68 Da and a cLogP of 0.65. The ligand carries two hydrogen-bond donors and four acceptors across three rotatable bonds, and it violates none of the Lipinski rules. These properties make clothianidin a tractable small-molecule target for ligand-conditioned protein binder design: it is small enough to sit fully within a designed pocket, polar enough to reward complementary hydrogen-bonding and charge-pairing interactions, and rigid enough that a single dominant conformer can anchor the design calculation.

The target-level QED drug-likeness score is 0.3552, which reflects the molecule's polar surface and modest lipophilicity rather than any failure of developability. The synthetic accessibility score is 0.7367 on the SAScore scale where 1 is easy and 10 is hard, placing clothianidin firmly in the readily synthesizable range. The ADMET aggregate risk score is 0.5634, with a toxicity risk component of 0.9597, a high value that is consistent with the compound's intended insecticidal activity and its known action on nicotinic acetylcholine receptors. These ligand-level properties are computed once for the target and frame the entire campaign: the designed binders must recognize a small, polar, synthetically accessible molecule whose toxicity profile is a feature of the target, not a liability of the binder.

Ligand Preparation

The campaign began with RDKit conformer generation and partial charge assignment for clothianidin. The ligand's three rotatable bonds define a limited conformational space, and the chlorothiazole ring provides a rigid aromatic anchor. The nitroimine group, with its delocalized negative charge, and the guanidine-like donor region create a strongly directional electrostatic surface. The CCD code CT4 identifies the ligand in the structural pipeline, and the preparation stage produced the conformer set and charge model that fed directly into pharmacophore fingerprinting and binder generation.

The prepared ligand structure served as the single-site design interface. Because clothianidin is the binding interface itself, the pipeline did not perform patch discovery on a protein surface. Instead, the ligand's prepared geometry defined the complementary surface that the generator was asked to build. This is the core constraint of the single-site design stage: the protein binder must wrap around the ligand, not the other way around.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified seven features on clothianidin. The feature inventory breaks down as follows:

• Acceptor.SingleAtomAcceptor: 1
• Aromatic.Arom5: 1
• Donor.SingleAtomDonor: 2
• Hydrophobe.ThreeWayAttach: 1
• LumpedHydrophobe.Nitro2: 1
• PosIonizable.Guanidine: 1

The single aromatic feature corresponds to the chlorothiazole ring, while the two donor features map to the guanidine-like nitrogen atoms. The nitro group contributes both the LumpedHydrophobe.Nitro2 feature and, through its oxygen atoms, part of the acceptor character. The PosIonizable.Guanidine feature is particularly important for binder design: it marks a region where a designed protein can place a negatively charged residue, such as aspartate or glutamate, to form a salt bridge. The Hydrophobe.ThreeWayAttach feature captures the methylene carbon linking the thiazole to the guanidine chain, a small hydrophobic patch that can be buried against aliphatic protein side chains.

These seven features were encoded into Morgan fingerprints, MXFP descriptors, and GNN embeddings via molfeat. The resulting pharmacophore representation conditioned the binder generator, telling it where to place hydrogen-bonding residues, where to build hydrophobic contacts, and where to position a complementary negative charge.

Binder Generation

The campaign generated 101 candidates using the AffiniBind pipeline version 0.5.0. The source-tool split is clean: 20 candidates came from BoltzGen alone, and 81 came from BoltzGen followed by proofreading. All 101 candidates carry a generator confidence score, and 20 candidates passed the generator prefilter. The prefilter excluded zero candidates, which means the 81 proofread candidates joined the table after the prefilter stage as variant or union rows that were never screened by that prefilter. This is an important structural detail: the prefilter operated on the 20 BoltzGen parents, and the proofread variants were added downstream without passing through that gate.

The generator produced a range of confidence scores. The median generator confidence for the 20 BoltzGen candidates is 0.5829, and the median for the 81 BoltzGen plus proofread candidates is 0.5806. The two source tools therefore produced comparable confidence distributions, with the proofread set slightly lower at the median but spanning a wider range of variants. The BoltzGen candidates also carry median developability metrics: solubility score 0.7984, SAP3D spatial score -1.1464, average packing density 15.2551, and flexible loop fraction 0.4089. These medians describe the parent generation before proofreading, and they establish the baseline quality of the designed binders.

Top Candidates

The ranked table below lists eight candidate rows representing four distinct designs. Six of the eight rows are variants of two parent designs, and the ranking reflects the parent's generator confidence score propagated to its variants. The two parent designs are BG_SM_DENOVO_CT4_E2E_boltzgen_binder_011 and BG_SM_DENOVO_CT4_E2E_boltzgen_binder_001, with the former contributing five proofread variants and the latter contributing one. The eighth row is a proofread variant of a third parent, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_019.

The table carries the two parent designs that have full metric coverage. The top-ranked parent, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_011, has a generator confidence score of 0.6964, a docking score of -2.63 kcal/mol, a solubility score of 0.8736, a SAP3D spatial score of -1.1925, an average packing density of 16.7013, and a flexible loop fraction of 0.4026. Its five proofread variants, named with the __pf1_00 through __pf1_04 suffixes, share the parent's source tool designation of boltzgen+proofread and the refold structure source, but their individual metric values are not part of the article's data surface.

The second parent, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_001, has a generator confidence score of 0.6955, a docking score of -4.10 kcal/mol, a solubility score of 0.8140, a SAP3D spatial score of -1.3111, an average packing density of 15.9245, and a flexible loop fraction of 0.4151. This design docks more favorably than the top-ranked parent by 1.47 kcal/mol, despite a marginally lower generator confidence. The difference between the two parents is instructive: the generator's confidence ranking does not perfectly track docking score, and the second parent's stronger docking performance comes with a slightly lower solubility score and a slightly higher flexible loop fraction.

The eighth listed row, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_019__pf1_03, is a proofread variant of a third parent design. Its source tool is boltzgen+proofread and its structure source is refold. The parent design BG_SM_DENOVO_CT4_E2E_boltzgen_binder_019 is not among the listed rows, and the variant's own metric values are limited to its source tool and structure source designations.

The campaign's best docking score is -4.58 kcal/mol, which appears in the campaign summary as the best value across all 101 candidates. The best QED score is 0.355, which is the target-level ligand property, not a per-candidate value. The best ADMET risk is 0.563, again a target-level property. These campaign-level extremes frame the overall performance: the docking scores are modest, in the -2.6 to -4.6 kcal/mol range, and the ligand's QED and ADMET values are fixed properties of clothianidin itself.

Docking and Developability

Docking validation used AutoDock Vina with 10 modes per candidate, and the best mode was recorded for each scored candidate. The two parent designs in the table carry docking scores of -2.63 kcal/mol and -4.10 kcal/mol, respectively. The campaign's best docking score of -4.58 kcal/mol exceeds both of these parent values, indicating that a candidate outside the listed table achieved the strongest predicted binding. The docking scores are accompanied by GNINA CNN pose and affinity predictions for the parent designs: BG_SM_DENOVO_CT4_E2E_boltzgen_binder_011 has a CNN pose score of 0.4630 and a CNN affinity of 4.9560, while BG_SM_DENOVO_CT4_E2E_boltzgen_binder_001 has a CNN pose score of 0.7426 and a CNN affinity of 3.6530. The higher CNN pose score for the second parent is consistent with its more favorable docking score.

The developability metrics for the parent designs are strong. Both parents have solubility scores above 0.81, placing them in the soluble class. The SAP3D spatial scores are -1.1925 and -1.3111, both negative values that indicate favorable spatial packing. The average packing densities are 16.7013 and 15.9245, and the flexible loop fractions are 0.4026 and 0.4151. The protease-exposed motif counts are 1.0 for the first parent and 2.0 for the second, and both parents have zero buried unsatisfied polar atoms. These values describe well-packed, soluble designs with no obvious developability liabilities.

The generator also produced structural quality metrics. The first parent has a gen_iptm of 0.6516, a gen_plddt of 0.7866, and a gen_rmsd of 0.69 Å. The second parent has a gen_iptm of 0.6743, a gen_plddt of 0.8310, and a gen_rmsd of 0.985 Å. The gen_ligand_iptm values match the gen_iptm values, indicating that the ligand is well integrated into the designed interface. The gen_min_interaction_pae values are 3.1472 and 2.3039, and the gen_interaction_pae values are 9.9264 and 8.1012. The second parent's lower interaction PAE values suggest a more confident ligand placement. The gen_complex_iplddt values are 0.5543 and 0.6793, and the gen_complex_pde values are 0.8065 and 0.6824. The gen_ptm values are 0.8534 and 0.8375, with gen_target_ptm values of 0.8665 and 0.8382. Both parents have gen_protein_iptm values of 0.0000, which reflects the single-site design constraint where the ligand is the interface and the protein is built around it.

The refolded structures show one salt bridge for the first parent and one hydrogen bond for the second parent, as captured by the gen_plip_saltbridge_refolded and gen_plip_hbonds_refolded values. The gen_delta_sasa_refolded values are 300.8685 and 290.6339, indicating substantial burial of surface area upon binding. The gen_affinity_pred_value values are 0.4934 and 0.5684, with binary probability values of 0.3642 and 0.4257. These affinity predictions are modest, consistent with the docking scores in the -2.6 to -4.6 kcal/mol range.

The ADMET assessment for the campaign is anchored by the target-level values: QED 0.3552, ADMET aggregate risk 0.5634, and ADMET toxicity risk 0.9597. These are properties of clothianidin itself, not of the designed binders. The high toxicity risk score reflects the ligand's insecticidal mechanism, and it does not constrain the developability of the protein binders. The synthetic accessibility score of 0.7367 applies to the ligand, not to the protein candidates, and it confirms that clothianidin is readily available for experimental validation.

Validation Plan

The campaign's ordering stage produced order files for two planned experiments, both available for reproduction from the target's experiments tab. The first is a yeast surface display screen using IDT oPools oligo pools covering all 101 candidates, tiered by generator confidence, for gap-repair cloning into the pCTCON2 vector. The second is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 20 candidates, with the T7 promoter plus GG plus SD plus Kozak plus ATG plus FLAG construct context using the E. coli MFC codon table. These experiments are planned and ready to run; they have not been executed, and no Kd values or titration curves exist yet. The order files and protocol documents are downloadable from the experiments tab, making the campaign's validation stage reproducible by another lab.

Honest Assessment

The clothianidin campaign produced a workable set of ligand-conditioned binder candidates, but the results expose a clear tension between generator confidence and docking performance. The top-ranked parent by generator confidence, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_011, docks at -2.63 kcal/mol, while the second parent, BG_SM_DENOVO_CT4_E2E_boltzgen_binder_001, docks at -4.10 kcal/mol with a confidence score only 0.0009 lower. The campaign's best docking score of -4.58 kcal/mol sits outside the listed table, which means the ranking by generator confidence did not surface the strongest binder. A follow-up campaign should re-rank candidates by docking score after generation, or use docking as a post-generation filter before selecting candidates for experimental validation.

The pharmacophore analysis identified seven features, including a PosIonizable.Guanidine that should drive salt-bridge formation, yet the refolded structures show only one salt bridge for the first parent and one hydrogen bond for the second. The designed binders are not fully exploiting the ligand's charge and hydrogen-bonding potential. A follow-up campaign could bias the generator toward placing a negatively charged residue near the guanidine group and a hydrogen-bonding residue near the nitroimine oxygen, using the pharmacophore features as explicit constraints rather than passive conditioning.

The developability metrics are the campaign's strongest result. Both parent designs are soluble, well-packed, and free of buried unsatisfied polar atoms. The SAP3D spatial scores are negative, the packing densities are in the expected range, and the flexible loop fractions are moderate. These designs are likely to express and fold, which is the first hurdle for any binder campaign. The docking scores, however, are modest, and the affinity predictions are correspondingly cautious. The planned yeast surface display screen and IVTT TR-FRET assay will determine whether the predicted binding translates to measurable affinity, but the computational evidence suggests that the current designs are a starting point rather than a finished product. A follow-up campaign should focus on improving docking scores through iterative design around the pharmacophore features, with particular attention to the guanidine salt bridge and the nitro group's acceptor potential.