Designing Protein Binders Against PFNA: A Ligand-Conditioned 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

Perfluorononanoic acid (PFNA) is a nine-carbon perfluorinated carboxylic acid with the SMILES string O=C(O)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)F. The molecule carries a single carboxylic acid head group and a fully fluorinated tail, giving it a molecular weight of 464.07 Da and a computed cLogP of 5.08. PFNA is a persistent environmental contaminant and a member of the perfluoroalkyl substance family, which makes it a relevant target for engineered recognition elements. The CCD code for PFNA is 4I6.

From a medicinal chemistry perspective, PFNA is a challenging ligand. It has one hydrogen bond donor and one hydrogen bond acceptor, both contributed by the carboxylic acid. The molecule has seven rotatable bonds, but the fluorinated chain is conformationally constrained by fluorine-fluorine repulsion and the helical twist typical of perfluorinated alkanes. The ligand violates one Lipinski rule, driven by its high lipophilicity and molecular weight. Its QED drug-likeness score is 0.5343, and its synthetic accessibility score is 0.8122 on the standard SAScore scale where 1 is easy and 10 is hard. The ADMET aggregate risk score for PFNA is 0.4067, with a toxicity risk score of 0.6761.

These ligand properties matter because they define the chemical surface a designed protein must recognize. The perfluorinated tail offers a large hydrophobic contact area but no polar handles beyond the terminal carboxylate. A successful binder must therefore exploit shape complementarity and the single ionizable group, while tolerating the ligand's low drug-likeness and elevated toxicity risk.

Ligand Preparation

The campaign began with RDKit conformer generation and partial charge assignment for PFNA. The ligand was prepared as a single molecular entity from its SMILES string, with the CCD code 4I6 used as the structural reference. The preparation stage produced the conformer ensemble used for pharmacophore fingerprinting and for the subsequent single-site design stage.

The ligand's computed properties reflect its perfluorinated character. The molecular weight of 464.07 Da places PFNA well above the typical small molecule range for high-affinity drug-like binders, and the cLogP of 5.08 indicates strong preference for nonpolar environments. The single carboxylic acid group is the only ionizable functionality, and at physiological pH it exists predominantly as the carboxylate anion. This anionic character is captured in the pharmacophore analysis described below.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified four features for PFNA. The feature counts are:

The two single-atom acceptor features correspond to the two oxygen atoms of the carboxylic acid group. The single donor feature corresponds to the acidic hydrogen of the carboxylic acid. The acidic group feature captures the ionizable carboxylate, which is the dominant electrostatic handle on the ligand. The perfluorinated tail contributes no pharmacophore features in this scheme, which is expected: the carbon-fluorine bonds are not classified as hydrogen bond acceptors or donors by the feature definitions used here.

This pharmacophore profile is sparse. A designed binder must therefore rely heavily on shape complementarity and van der Waals contacts with the fluorinated chain, while using the carboxylate as the primary polar anchor. The single-site design stage treated the ligand as the binding interface, with no patch discovery step.

Binder Generation

The campaign generated 194 candidates using the AffiniBind pipeline version 0.5.0. The source-tool split is 80 candidates from boltzgen and 114 candidates from boltzgen+proofread. All 194 candidates carry a generator confidence score. The generator prefilter accepted 80 candidates and excluded 0 candidates. The 114 candidates from the boltzgen+proofread source joined the table after the prefilter stage as variant or union rows.

The median generator confidence score for the 80 boltzgen candidates is 0.8108. The median for the 114 boltzgen+proofread candidates is 0.8777. The proofread step therefore shifts the confidence distribution upward, which is consistent with its role as a refinement pass over the initial boltzgen designs.

The boltzgen candidates also carry median developability metrics. The median solubility score is 0.7657, the median SAP3D spatial score is -1.2109, the median average packing density is 15.2645, and the median flexible loop fraction is 0.3056. These medians describe the 80 scored boltzgen candidates and provide a baseline for comparing individual designs.

Top Candidates

The ranked table below lists eight candidate rows from the campaign. These eight rows represent five distinct designs: three parent designs and five variant rows derived from two of those parents. The ranking reflects the generator confidence score of each parent design propagated to its variants. The variants named with a parent= design are proofread versions of the parent, and they carry their own source tool label of boltzgen+proofread.

The top-ranked parent design, BG_SM_DENOVO_4I6_L80_E2E_boltzgen_binder_041, carries a generator confidence score of 0.9549 and a docking score of -9.50 kcal/mol. Its solubility score is 0.6630, placing it in the soluble class. The SAP3D spatial score is -1.6498, and the average packing density is 15.0303. The flexible loop fraction is 0.1970, the lowest among the three parent designs shown here. The design has zero protease-exposed motifs and zero buried unsatisfied polar atoms. Its refolded structure shows one hydrogen bond to the ligand and a delta SASA of 454.46 Ų upon complex formation. The generator's ligand iPTM is 0.8982, and the interaction PAE is 5.9318.

The second parent design, BG_SM_DENOVO_4I6_L80_E2E_boltzgen_binder_042, has a generator confidence score of 0.9515 and a docking score of -7.70 kcal/mol. Its solubility score is 0.6636, also in the soluble class. The SAP3D spatial score is -1.2550, and the average packing density is 15.3699. The flexible loop fraction is 0.2466. This design has one protease-exposed motif and zero buried unsatisfied polar atoms. Its refolded structure shows zero hydrogen bonds to the ligand and a delta SASA of 393.75 Ų. The generator's ligand iPTM is 0.9048, and the interaction PAE is 5.6150.

The third parent design, BG_SM_DENOVO_4I6_L80_E2E_boltzgen_binder_021, has a generator confidence score of 0.9338 and a docking score of -8.78 kcal/mol. Its solubility score is 0.8111, the highest among the three parents. The SAP3D spatial score is -1.9607, and the average packing density is 14.7879. The flexible loop fraction is 0.2727. This design has four protease-exposed motifs and zero buried unsatisfied polar atoms. Its refolded structure shows one hydrogen bond to the ligand and a delta SASA of 362.72 Ų. The generator's ligand iPTM is 0.8935, and the interaction PAE is 6.2911.

The five variant rows derived from these parents are proofread versions. The variants of binder_041 are __pf1_01, __pf1_00, and __pf1_02. The variants of binder_042 are __pf1_02 and __pf1_00. These variants carry the boltzgen+proofread source tool label and a structure_source of refold. Their ranking positions follow the parent confidence ordering, with the binder_041 variants occupying ranks 2 through 4 and the binder_042 variants occupying ranks 5 and 6 in the full eight-row listing.

Docking and Developability

The docking stage used AutoDock Vina with 10 modes per candidate. The best docking score across the campaign is -10.61 kcal/mol. Among the three parent designs shown in the table, the docking scores are -9.50 kcal/mol for binder_041, -7.70 kcal/mol for binder_042, and -8.78 kcal/mol for binder_021. The GNINA CNN pose confidence scores for these three parents are 0.7261, 0.9368, and 0.4569 respectively. The GNINA CNN affinity values are 5.1810, 6.6680, and 4.8850 respectively.

The developability metrics for the parent designs show a consistent picture. All three parents are classified as soluble, with solubility scores of 0.6630, 0.6636, and 0.8111. The SAP3D spatial scores are all negative, ranging from -1.9607 to -1.2550, which indicates favorable spatial packing. The average packing densities range from 14.7879 to 15.3699. The flexible loop fractions range from 0.1970 to 0.2727.

The generator-level metrics provide additional confidence in the designs. The ligand iPTM values are 0.8982, 0.9048, and 0.8935 for the three parents. The interaction PAE values are 5.9318, 5.6150, and 6.2911. The complex ipLDDT values are 0.5757, 0.6449, and 0.5023. The complex PDE values are 0.4586, 0.4450, and 0.4903. These metrics describe the predicted quality of the protein-ligand interface and the overall complex.

The affinity prediction values from the generator are 0.8666, 0.8252, and 1.0963 for the three parents. The binary affinity probabilities are 0.1483, 0.1705, and 0.1615. These values reflect the generator's internal assessment of binding likelihood, and they should be interpreted alongside the docking scores rather than as independent affinity measurements.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental validation routes. These experiments are available for reproduction and have not been run.

The first planned experiment is a yeast surface display screen using IDT oPools oligo pools covering all 194 candidates. The candidates are tiered by confidence for gap-repair cloning into the pCTCON2 vector. This screen is designed to test binding of the displayed proteins against PFNA in a high-throughput format.

The second planned experiment is an IVTT TR-FRET assay using 96-well IDT gBlock constructs covering 80 candidates. The constructs use the T7 promoter with a GG + SD + Kozak + ATG + FLAG construct context, using the E. coli MFC codon table. This assay is designed to test binding in a cell-free format with fluorescence readout.

The order files and protocol documents are downloadable from the target's experiments tab. These resources are ready to run for any lab that wants to reproduce or extend the campaign.

Honest Assessment

This campaign against PFNA produced a workable set of binder candidates with a clear top design. The best docking score of -10.61 kcal/mol across the full campaign is encouraging for a ligand with only one polar handle and a large perfluorinated surface. The top parent design, binder_041, combines a generator confidence of 0.9549 with a docking score of -9.50 kcal/mol and a soluble classification. The proofread variants of this parent occupy the next three ranking positions, which suggests the refinement step preserved the parent's favorable properties.

The pipeline's strengths are evident in the consistency of the developability metrics. All three parent designs shown here are soluble, with negative SAP3D spatial scores and packing densities near 15. The buried unsatisfied polar count is zero for all three parents, which indicates the designs avoid internal polar strain. The flexible loop fractions are moderate, ranging from 0.1970 to 0.2727, and the protease-exposed motif counts are low for two of the three parents.

The main challenge is the ligand itself. PFNA's QED of 0.5343 and ADMET toxicity risk of 0.6761 reflect its perfluorinated character, and these properties constrain what a binder can achieve in a therapeutic context. The pharmacophore analysis identified only four features, all concentrated in the carboxylic acid head group. The perfluorinated tail offers no hydrogen bonding or ionizable features, so the designs must rely on shape complementarity and hydrophobic contacts. The docking scores and GNINA pose confidence values suggest the designs do engage the ligand, but the affinity prediction probabilities from the generator are modest, ranging from 0.1483 to 0.1705 for the three parents.

A follow-up campaign would benefit from targeting the carboxylate more explicitly. The current designs show one hydrogen bond to the ligand in two of the three parents, and zero in the third. Increasing the number of polar contacts to the carboxylic acid, perhaps through a designed arginine or lysine side chain, could improve both docking scores and affinity predictions. The perfluorinated tail is a large hydrophobic surface, and the designs already show substantial delta SASA values upon complex formation, but the packing density and loop fraction metrics suggest there is room to tighten the interface.

The validation plan is ready to run, and the yeast surface display screen covering all 194 candidates is the logical next step. The IVTT TR-FRET assay covering 80 candidates provides a complementary cell-free readout. Both experiments are available for reproduction, and the order files are downloadable from the experiments tab. The campaign's computational results justify moving to experimental validation, with the caveat that PFNA's toxicity and low drug-likeness will require careful interpretation of any binding data in a translational context.