Designing Protein Binders for PFHpA: 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.

Perfluoroheptanoic acid (PFHpA) belongs to the perfluoroalkyl carboxylic acid family, a class of persistent environmental contaminants that resist metabolic breakdown and accumulate in biological systems. The molecule presents a carboxylic acid head group attached to a six-carbon perfluorinated tail, giving it both a polar anchor point and a highly hydrophobic fluorocarbon surface. That combination makes PFHpA a demanding target for protein binder design: the ligand offers only one hydrogen bond donor and one acceptor, yet its fluorinated tail creates a large, low-polarity surface that a complementary protein pocket must wrap tightly to achieve specificity.

The campaign described here used the AffiniBind platform, pipeline version 0.5.0, to generate and rank 195 candidate protein binders against PFHpA. The pipeline ran through ligand preparation, pharmacophore fingerprinting, single-site design, binder generation, and docking plus ADMET assessment. The best docking score across the campaign reached -9.23 kcal/mol, and the top-ranked candidate carried a generator confidence score of 0.9374.

The Target Ligand

PFHpA is a seven-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)F and the CCD code 4EI. The molecule has a molecular weight of 364.06 Da and a computed cLogP of 3.81. Its single carboxylic acid group provides one hydrogen bond donor and one hydrogen bond acceptor, while the perfluorinated chain contributes five rotatable bonds and a substantial hydrophobic surface.

The ligand's drug-likeness profile is surprisingly workable for a perfluorinated acid. The computed QED score is 0.7539, with zero Lipinski violations. The synthetic accessibility score is 0.8144 on the standard SAScore scale where 1 is easy and 10 is hard, indicating that PFHpA itself is readily accessible as a starting material. The ADMET aggregate risk score is 0.3805, with a toxicity risk component of 0.7186. That toxicity risk is the dominant concern for this ligand, consistent with the known persistence and bioaccumulation behavior of perfluoroalkyl acids.

Ligand Preparation

The ligand preparation stage began with the PFHpA SMILES string and the CCD code 4EI. RDKit conformer generation produced the three-dimensional starting geometry, and partial charges were assigned to prepare the ligand for downstream docking and generator conditioning. The prepared ligand carried the target-level properties listed above: molecular weight 364.06 Da, cLogP 3.81, one hydrogen bond donor, one hydrogen bond acceptor, five rotatable bonds, and zero Lipinski violations.

The preparation stage also computed the ligand's QED drug-likeness score of 0.7539 and its synthetic accessibility score of 0.8144. These values are target-level facts, computed once for PFHpA and used throughout the campaign as the reference point for drug-likeness and ADMET assessment. The ADMET aggregate risk score of 0.3805 and the toxicity risk score of 0.7186 were likewise computed at this stage and apply to the ligand itself, not to individual protein candidates.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified four features on PFHpA that the binder generator would need to satisfy. The feature breakdown is:

The two single-atom acceptor features correspond to the carboxylic acid oxygens, while the single donor feature is the acidic proton. The acidic group feature captures the deprotonation potential of the carboxylic acid at physiological pH. These four features define the polar interaction surface that a successful binder must engage. The fluorinated tail contributes no pharmacophore features in this fingerprinting scheme, which means the generator had to rely on shape complementarity and hydrophobic packing to recognize that portion of the ligand.

Binder Generation

The binder generation stage produced 195 candidates total. Of these, 80 candidates came from the boltzgen source tool, and 115 came from boltzgen+proofread. The generator prefilter accepted 80 candidates and excluded zero, while 115 candidates joined the table after the prefilter stage as variant or union rows that were never screened by the prefilter.

The per-tool median metrics reveal a meaningful difference between the two source tools. The boltzgen candidates, with 80 scored designs, achieved a median generator confidence score of 0.7562, a median solubility score of 0.7837, a median SAP3D spatial score of -1.1828, a median average packing density of 15.5115, and a median flexible loop fraction of 0.3080. The boltzgen+proofread candidates, with 115 scored designs, achieved a median generator confidence score of 0.8728. The proofread step raised the median confidence by roughly 0.12 points, suggesting that the proofreading pass systematically improved the generator's own assessment of its outputs.

Top Candidates

The ranked candidate table lists eight rows representing three distinct designs. Seven of the eight rows are variants of two parent designs: six variants descend from BG_SM_DENOVO_4EI_L80B_E2E_boltzgen_binder_061, and two variants descend from BG_SM_DENOVO_4EI_L80B_E2E_boltzgen_binder_062. The ranking reflects the parent design's generator confidence score propagated to its variants, so the variant cluster around binder_061 occupies the top six positions.

The parent design BG_SM_DENOVO_4EI_L80B_E2E_boltzgen_binder_061 carries the campaign's top generator confidence score of 0.9374. Its full metric profile is the richest in the table:

The parent design's docking score of -8.68 kcal/mol places it among the stronger binders in the campaign, though it is not the single best docking score. The GNINA CNN pose confidence of 0.8371 indicates that the docked pose is geometrically plausible, and the CNN affinity value of 5.8790 reflects the network's internal affinity estimate. The refolded structure shows one hydrogen bond to the ligand and zero salt bridges, with a delta SASA of 401.55 Ų, indicating substantial burial of surface area upon complex formation.

The six proofread variants of binder_061 (__pf1_00 through __pf1_04) and the two proofread variants of binder_062 (__pf1_01 and __pf1_02) each carry the source tool label boltzgen+proofread and the structure source refold. These variant rows represent the proofreading pass applied to the parent designs, and their presence in the ranked table reflects the parent's confidence score propagated through the variant cluster.

Docking and Developability

The docking stage used AutoDock Vina with GNINA CNN rescoring. The best docking score across the entire campaign reached -9.23 kcal/mol, which is stronger than the top-ranked parent's -8.68 kcal/mol. This gap between the best docking score and the top-ranked candidate's docking score is worth noting: the generator confidence ranking and the docking score ranking do not perfectly align, and a follow-up campaign might weight docking score more heavily in the final ranking.

The developability metrics for the top parent design are encouraging. The solubility score of 0.6993 places the design in the soluble class. The SAP3D spatial score of -1.5641 is negative, which is the favorable direction for this metric. The average packing density of 15.6104 and the flexible loop fraction of 0.2597 indicate a reasonably well-packed core with a moderate proportion of flexible loop regions. The buried unsatisfied polar count of 0.0000 is a strong result: no polar atoms are buried without hydrogen bonding partners, which is a common failure mode in de novo designed binders.

The protease exposed motif count of 1.0000 flags one potential developability concern. A single exposed protease-sensitive motif may be acceptable for a research reagent, but it would need attention if this binder were advanced toward therapeutic or diagnostic applications where proteolytic stability matters.

The ADMET assessment for this campaign is target-level, computed once for PFHpA itself. The ligand's ADMET aggregate risk score of 0.3805 and toxicity risk score of 0.7186 are the relevant values for understanding the drug-likeness context of this campaign. The high toxicity risk component is consistent with the known persistence and bioaccumulation of perfluoroalkyl acids, and it means that even a high-affinity binder would face significant hurdles in any therapeutic application. The QED score of 0.7539 and zero Lipinski violations indicate that PFHpA itself is not a problematic small molecule from a drug-likeness perspective, but the toxicity risk dominates the ADMET picture.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental validation routes. These experiments are available and ready to run, but they have not been executed.

The first planned experiment is a yeast surface display screen using IDT oPools oligo pools covering all 195 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen would provide a direct experimental readout of binding to PFHpA across the full candidate set.

The second planned experiment is an IVTT TR-FRET assay using 96-well IDT gBlock constructs covering 80 candidates. The construct context includes the T7 promoter, GG, SD, Kozak, ATG, and FLAG tag, using the E. coli MFC codon table. This assay would provide quantitative binding measurements in a cell-free format.

The order files and protocol documents are downloadable from the target's experiments tab. Both experiments are available for another lab to reproduce, and the tiered confidence ordering means the highest-confidence candidates can be tested first.

Honest Assessment

This campaign produced a substantial candidate set against a challenging fluorinated ligand, and the pipeline's outputs are internally consistent enough to justify experimental follow-up. The generator confidence scores are strong, with the top parent design reaching 0.9374 and the proofread median at 0.8728. The docking scores are in a reasonable range for de novo designed binders, with the campaign best at -9.23 kcal/mol and the top-ranked parent at -8.68 kcal/mol. The developability metrics for the top parent are mostly favorable: soluble class, zero buried unsatisfied polars, and a negative SAP3D spatial score.

The pipeline struggled most visibly with the affinity prediction signal. The top parent's affinity probability of 0.0896 is low, and this sits awkwardly alongside the strong docking score and high generator confidence. The single protease exposed motif on the top parent is another flag that would need attention in a follow-up campaign. The variant cluster structure also means the ranked table overrepresents two parent designs: seven of the eight listed rows descend from just two parents, and the ranking reflects parent confidence propagated to variants rather than independent candidate quality.

A follow-up campaign would target several changes. First, the ranking should weight docking score and affinity prediction more heavily alongside generator confidence, since the current confidence-first ranking placed a candidate with a -8.68 kcal/mol docking score above other candidates that may have docked better. Second, the proofread step's consistent confidence boost suggests it should be applied more broadly, but its variants need their own docking and developability measurements rather than inheriting the parent's ranking position. Third, the protease exposed motif on the top parent should be engineered out in the next round. Finally, the planned YSD and TR-FRET experiments should be run as soon as practical, because the gap between the generator's confidence and the affinity model's skepticism can only be resolved by experimental binding data.