Designing Protein Binders Against 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.

The Target Ligand

Perfluoroheptanoic acid (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. It carries the CCD code 4EI. The molecule belongs to the perfluoroalkyl acid family, a class of persistent environmental contaminants that resist metabolic breakdown because of the strength of the carbon-fluorine bond and the electron-withdrawing character of the fully fluorinated tail. PFHpA itself is a shorter-chain relative of the more widely studied PFOA, and its accumulation in water sources and biological tissues has made selective recognition a practical problem for detection and remediation workflows.

From a protein design perspective, PFHpA presents a distinctive interface challenge. The ligand is amphiphilic in an unusual way: a single carboxylic acid head group provides the only conventional hydrogen-bond donor and acceptor functionality, while the perfluorinated tail offers a hydrophobic surface that is simultaneously lipophobic and oleophobic. Fluorinated surfaces interact weakly with hydrocarbon side chains, so a naive hydrophobic pocket built from leucine or isoleucine will not necessarily engage the tail productively. The design task is therefore to build a pocket that complements the fluorinated chain through shape and dispersion while anchoring the carboxylate with precise polar contacts. The campaign described here used the AffiniBind small molecule pipeline, version 0.5.0, to generate and rank 115 candidate binders against this target.

Ligand Preparation

The ligand preparation stage began with the PFHpA SMILES string and generated a three-dimensional conformer set using RDKit, followed by partial charge assignment. The target-level properties computed once for PFHpA place it in an interesting region of chemical space for a small molecule target. The molecular weight is 364.06 Da, and the computed cLogP is 3.81. The molecule has one hydrogen-bond donor, one hydrogen-bond acceptor, and five rotatable bonds. It records zero Lipinski violations, which is expected for a compound of this size and polarity profile.

The drug-likeness metrics tell a more nuanced story. The QED score for PFHpA is 0.7539, which sits comfortably in the attractive range by the usual thresholds, but the ADMET aggregate risk score is 0.3805, and the ADMET toxicity risk score climbs to 0.7186. That toxicity risk is the dominant concern for this ligand as a target, and it reflects the known persistence and bioaccumulation liabilities of perfluorinated acids rather than any acute reactivity problem. 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 straightforward to obtain or synthesize for assay development. These ligand-level properties are target facts, not candidate facts, and they frame the downstream filtering: a binder designed against PFHpA must be judged in part by whether it can be produced and deployed as a detection reagent, not by whether the ligand itself is a drug lead.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified four features on PFHpA. The feature counts break down as two single-atom acceptors, one single-atom donor, and one acidic group from the ionizable carboxylate. This is a compact pharmacophore, and it accurately reflects the chemical reality of the molecule: the carboxylic acid head group accounts for all three polar features, while the perfluorinated tail contributes no conventional hydrogen-bonding or ionizable functionality.

That distribution of features has direct consequences for binder design. A protein pocket that recognizes PFHpA must concentrate its polar contacts at one end of the ligand, around the carboxylate, while the remainder of the interface must rely on shape complementarity and dispersion interactions with the fluorinated chain. The pharmacophore fingerprint does not capture fluorophilic contacts, which are not standard hydrogen-bond or ionic features, but it does tell the generator where the anchor points are. The single acidic group is particularly important: at physiological pH the carboxylate will be deprotonated, so a complementary pocket should place a positive charge or a hydrogen-bond donor near that position.

Binder Generation

The binder generation stage produced 115 candidates in total. The source-tool split shows 20 candidates from the boltzgen generator and 95 from the boltzgen+proofread combination. All 115 candidates carry a generator confidence score, and 20 candidates passed the generator prefilter. The prefilter excluded zero candidates, which means the 95 boltzgen+proofread rows 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 for interpreting the ranking: the campaign's candidate count is 115, but the prefiltered set is 20, and the remaining 95 are proofread variants of designs that entered through the boltzgen route.

The per-tool medians give a sense of how the two source groups behave. For the 20 boltzgen candidates, the median generator confidence score is 0.7703, the median solubility score is 0.7668, the median SAP3D spatial score is -1.2493, the median average packing density is 15.3072, and the median flexible loop fraction is 0.3391. For the 95 boltzgen+proofread candidates, the median generator confidence score is 0.7716. The proofread variants therefore sit very close to the boltzgen-only set in generator confidence, which is consistent with their origin as refinements of the same parent designs rather than independent de novo outputs.

Top Candidates

The ranked table below shows the eight candidate rows that carry the highest generator confidence scores in this campaign. These eight rows represent three distinct designs: one parent design with six variant rows, one parent design with two variant rows, and the parent row itself. The ranking reflects the parent's score propagated to its variants, so the top six rows are all variants of the same parent design, BG_SM_DENOVO_4EI_E2E_boltzgen_binder_002, and the seventh and eighth rows are variants of BG_SM_DENOVO_4EI_E2E_boltzgen_binder_003.

The top design, BG_SM_DENOVO_4EI_E2E_boltzgen_binder_002, carries a generator confidence score of 0.9166 and a docking score of -7.84 kcal/mol. Its GNINA CNN pose confidence is 0.7552, and the GNINA CNN affinity prediction is 5.1810. The interface metrics for this design are consistent with a well-packed binding site: gen_iptm is 0.8697, gen_plddt is 0.6296, and gen_rmsd is 0.6636 Å. The interaction PAE is 6.1843, and the minimum interaction PAE is 0.9827, indicating that the ligand is placed with high confidence relative to the designed pocket. The complex ipLDDT is 0.5423, and the complex PDE is 0.5501. The refolded structure shows one hydrogen bond to the ligand, zero salt bridges, and a delta SASA of 360.43 Ų upon binding. The affinity prediction value is 1.8078 with a binary probability of 0.0874.

The second distinct design, BG_SM_DENOVO_4EI_E2E_boltzgen_binder_003, scores slightly lower on generator confidence at 0.8908 and docks at -7.81 kcal/mol. Its GNINA CNN pose confidence drops to 0.5335, and the CNN affinity prediction is 4.7620. The interface metrics are somewhat different in character: gen_iptm is 0.8315, gen_plddt is 0.6858, and gen_rmsd is 0.5084 Å. The interaction PAE is 6.0504, and the minimum interaction PAE is 0.9615. The complex ipLDDT is 0.6024, and the complex PDE is 0.5515. The refolded structure for this design shows zero hydrogen bonds to the ligand and zero salt bridges, with a delta SASA of 293.96 Ų. The affinity prediction value is 1.6702 with a binary probability of 0.1409.

Docking and Developability

The docking stage used AutoDock Vina with 10 modes per candidate, and the best mode was mode 1 for every listed candidate. The best docking score across the campaign is -9.32 kcal/mol, which is stronger than the -7.84 kcal/mol recorded by the top-ranked design in the table. This is an important distinction: the table is ranked by generator confidence score, not by docking score, and the campaign's best docking result belongs to a candidate outside the top-ranked cluster shown here. The top-ranked design's docking score of -7.84 kcal/mol is still a solid result for a de novo binder against a small fluorinated ligand, but it is not the campaign's docking optimum.

The developability metrics for the top parent design are encouraging. The solubility score is 0.6941, which places the design in the soluble class. The SAP3D spatial score is -1.1714, and the average packing density is 14.7761. The flexible loop fraction is 0.2687, which suggests a reasonably ordered core with some conformational flexibility at the periphery. The protease-exposed motif count is 0.0000, and the buried unsatisfied polar count is 0.0000, both of which are favorable for stability and expression. The raw chain break count is 0.0000, the maximum consecutive C-N-A value is 1.4200, and the unplaced atom count is 0.0000, indicating that the refolded structure is chemically complete and topologically intact.

The ligand-level ADMET profile is the more sobering part of the picture. PFHpA's ADMET aggregate risk score is 0.3805, and its toxicity risk score is 0.7186. These are target-level facts, not candidate-level facts, and they reflect the inherent persistence and bioaccumulation concerns associated with perfluorinated acids. A binder designed against PFHpA is not being developed as a therapeutic; it is being developed as a recognition element for detection, sequestration, or environmental monitoring. In that context, the ligand's toxicity risk is a handling consideration for assay development rather than a disqualifying property of the binder itself.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental workflows, both available for reproduction from the target's experiments tab. These experiments have not been run; they are ready to run and available for another lab to reproduce.

The first planned experiment is a yeast surface display screen using IDT oPools oligo pools covering all 115 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen is designed to test binding of the displayed binders against labeled PFHpA in a high-throughput format. The tiering by confidence means the highest-scoring candidates from the generator ranking will be interrogated first, with the full 115-candidate pool available for deeper screening.

The second planned experiment is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 20 candidates. The constructs carry the T7 promoter plus GG plus SD plus Kozak plus ATG plus FLAG construct context, using the E. coli MFC codon table. This cell-free assay format is intended to measure binding in solution without the complications of surface display, providing an orthogonal readout to the yeast screen. The 20-candidate count matches the number of candidates that passed the generator prefilter, so this assay targets the prefiltered set specifically.

Honest Assessment

This campaign produced a coherent set of binder candidates against PFHpA, with a clear top design and a defensible ranking. The strongest result is the parent design BG_SM_DENOVO_4EI_E2E_boltzgen_binder_002, which combines a generator confidence score of 0.9166 with a docking score of -7.84 kcal/mol, a GNINA CNN pose confidence of 0.7552, and a refolded structure that places one hydrogen bond on the ligand. The developability profile for this design is clean: soluble class, zero buried unsatisfied polars, zero protease-exposed motifs, and zero chain breaks. The second distinct design, BG_SM_DENOVO_4EI_E2E_boltzgen_binder_003, is a reasonable backup with a slightly lower confidence score of 0.8908 and a docking score of -7.81 kcal/mol, but its refolded structure shows zero hydrogen bonds to the ligand, which is a meaningful weakness for a target whose only polar anchor is a single carboxylate.

The campaign's structural limitation is the clustering of the top-ranked rows. Six of the eight listed candidates are variants of a single parent design, and the remaining two are variants of a second parent. The ranking therefore reflects the parent's score propagated to its variants, not eight independent top candidates. The campaign as a whole generated 115 candidates, and the best docking score of -9.32 kcal/mol belongs to a candidate outside the top-ranked cluster shown in the table. A follow-up campaign should look carefully at that -9.32 kcal/mol candidate and ask whether its generator confidence score can be improved through proofreading or whether its docking advantage reflects a genuinely different binding mode worth pursuing.

The pharmacophore analysis also points to a design tension that the current candidates only partially resolve. PFHpA has four pharmacophore features, all concentrated in the carboxylate head group, and the perfluorinated tail offers no conventional polar anchors. The top design places one hydrogen bond on the ligand, which is a start, but a stronger binder would likely need to engage the carboxylate with a more complete polar network, perhaps a salt bridge or a bidentate hydrogen-bonding arrangement. The current designs rely heavily on shape complementarity and burial of the fluorinated tail, which is necessary but may not be sufficient for high-affinity recognition.

The ADMET picture is dominated by the ligand's own toxicity risk score of 0.7186, which is a target-level fact that no binder design can change. For detection and remediation applications this is a handling concern, not a design failure, but it does mean that any experimental validation will need to manage PFHpA carefully. The planned yeast surface display and IVTT TR-FRET assays are the right next steps, and the fact that they are available for reproduction with downloadable order files and protocols makes this campaign a useful starting point for a lab that wants to test these binders experimentally. A follow-up campaign would benefit from a more aggressive polar engagement strategy at the carboxylate, a closer look at the -9.32 kcal/mol docking candidate, and a deliberate effort to diversify the top-ranked cluster beyond variants of a single parent design.