Pentachlorophenol Binder Design with AffiniBind

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

Pentachlorophenol is a fully chlorinated phenol: five chlorine substituents arranged around a single aromatic ring bearing one hydroxyl group. The molecule is compact, rigid, and heavily halogenated, which makes it an instructive test case for ligand-conditioned binder design. It presents a dense hydrophobic surface with a single hydrogen-bond donor and a single acceptor, both contributed by the phenolic oxygen. There are no rotatable bonds, so the ligand arrives at the binding interface as a fixed scaffold rather than a flexible small molecule that can adapt to a pocket.

The ligand properties computed once for this campaign place pentachlorophenol at a molecular weight of 266.34 Da with a cLogP of 4.66. The molecule has one hydrogen-bond donor, one hydrogen-bond acceptor, and zero rotatable bonds. It records zero Lipinski violations. Its QED drug-likeness is 0.5346, and its synthetic accessibility score is 0.8478 on the standard 1-to-10 scale where 1 is easiest. The ADMET aggregate risk score is 0.4776, while the ADMET toxicity risk score is 0.9625. That toxicity value is the dominant liability in the target profile and shapes how the downstream developability readout should be interpreted.

The design problem here is not to optimize the ligand itself. Pentachlorophenol is what it is: a rigid, electron-poor aromatic with a single polar edge. The challenge is to generate a protein surface that can recognize that exact shape and chemistry, bury the chlorinated ring productively, and satisfy the phenolic oxygen without relying on conformational flexibility from the small molecule.

Ligand Preparation

The campaign began with the ligand SMILES string Oc1c(Cl)c(Cl)c(Cl)c(Cl)c1Cl, corresponding to the CCD code PCI. RDKit conformer generation produced the three-dimensional starting geometry, and partial charges were assigned for downstream docking and interaction scoring. Because pentachlorophenol has no rotatable bonds, the conformer ensemble is effectively a single rigid pose family. The preparation stage therefore reduces to protonation state assignment, charge placement, and geometry optimization of the aromatic core and its chlorine substituents.

That rigidity is a useful constraint for the binder generation stage. The ligand cannot twist to accommodate a poorly shaped pocket, so the generated protein must match the ligand's fixed presentation of its hydrophobic and polar features. The phenolic oxygen is the only polar contact point, while the five chlorine atoms and the aromatic ring form a contiguous hydrophobic patch that dominates the molecular surface.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified nine features for pentachlorophenol. The feature-type counts break down as follows:

The pharmacophore profile is dominated by hydrophobic features. Five of the nine features are classified as Hydrophobe.ThreeWayAttach, which reflects the branching chlorination pattern around the ring. The single aromatic feature captures the six-membered ring, and the single donor and single acceptor features both map to the phenolic oxygen. This is a ligand whose recognition problem is mostly about shape complementarity and hydrophobic burial, with one polar edge that must be satisfied precisely.

Binder Generation

The campaign generated 109 candidates using the AffiniBind pipeline version 0.5.0. The source-tool split is clean: 20 candidates came from boltzgen alone, and 89 came from boltzgen with proofreading. All 109 candidates carry a generator confidence score. The generator prefilter passed 20 candidates and excluded zero, while 89 candidates joined the table after the prefilter stage as variant or union rows.

That distribution matters for how the ranking should be read. The 20 boltzgen candidates are the primary designs that passed the prefilter. The 89 boltzgen+proofread candidates are variants of those parent designs, produced by the proofreading stage. The top of the ranked list is therefore not eight independent designs; it is a small number of parent designs with their proofread variants clustered around them.

The per-tool medians show the expected pattern. For the 20 boltzgen candidates, the median generator confidence score is 0.7547, the median solubility score is 0.7963, the median SAP3D spatial score is -1.6179, the median average packing density is 15.5289, and the median flexible loop fraction is 0.2649. For the 89 boltzgen+proofread candidates, the median generator confidence score is 0.7445. The proofread variants sit slightly below the primary designs in median confidence, which is consistent with proofreading exploring local sequence space around a parent rather than generating fully independent solutions.

Top Candidates

The ranked candidate table lists eight rows representing four distinct designs. Six of those rows are variants of two parent designs. The ranking reflects the parent's generator confidence score propagated to its variants, so the cluster around the top parent appears as the top six rows even though those variants are not independent top candidates.

The table above includes only the two parent designs that carry the full metric set. The six variant rows in the ranked list carry their own source tool, ligand gate status, and structure source values, but the developability and docking metrics belong to the parent lines. The correct reading is that the campaign's top parent design is BG_SM_DENOVO_PCI_E2E_boltzgen_binder_009 with a generator confidence score of 0.8803 and a docking score of -4.36 kcal/mol. Its five proofread variants occupy ranks 2 through 6. The second parent, BG_SM_DENOVO_PCI_E2E_boltzgen_binder_010, sits at rank 7 with a generator confidence score of 0.8243 and a docking score of -3.20 kcal/mol, followed by one proofread variant at rank 8.

The top parent's developability profile is solid. Its solubility score is 0.6437 with a soluble classification. The SAP3D spatial score is -2.3692, and the average packing density is 15.0566. The flexible loop fraction is 0.2075, meaning roughly one-fifth of the designed chain is in flexible loop regions. The protease-exposed motif count is 1.0000, and the buried unsatisfied polar count is 0.0000, which is a favorable sign for fold stability.

The second parent shows a higher average packing density of 19.6216 and a lower flexible loop fraction of 0.1351. Its SAP3D spatial score is -1.6335, and its solubility score is 0.6620. The docking score of -3.20 kcal/mol is weaker than the top parent's -4.36 kcal/mol, but the generator confidence score of 0.8243 keeps it in the top tier.

Docking and Developability

Docking validation used AutoDock Vina with 10 modes per candidate. The best docking score across the campaign is -6.69 kcal/mol. The top-ranked parent design by generator confidence, BG_SM_DENOVO_PCI_E2E_boltzgen_binder_009, docks at -4.36 kcal/mol with a GNINA CNN pose score of 0.7144 and a GNINA CNN affinity value of 5.0860. Its best mode is mode 1. The second parent, BG_SM_DENOVO_PCI_E2E_boltzgen_binder_010, docks at -3.20 kcal/mol with a GNINA CNN pose score of 0.5206 and a GNINA CNN affinity value of 5.2570, also with best mode 1.

The generator-level structural metrics for the top parent are encouraging. The gen_iptm is 0.8111, the gen_plddt is 0.7076, and the gen_rmsd is 0.3848 Å. The gen_ligand_iptm matches the gen_iptm at 0.8111, and the gen_min_interaction_pae is 1.3165. The gen_complex_iplddt is 0.6293, and the gen_interaction_pae is 5.3480. The gen_complex_pde is 0.5512, the gen_ptm is 0.8293, and the gen_target_ptm is 0.9413. The gen_protein_iptm is 0.0000, which reflects the single-site design context where the ligand is the binding interface.

The refolded structure metrics show zero hydrogen bonds and zero salt bridges from PLIP, with a delta SASA of 311.7009 Ų for the top parent. The affinity prediction value is 0.4744 with a binary probability of 0.4271. The raw chain break count is 0.0000, the maximum consecutive C-N distance is 1.3900 Å, and there are zero unplaced atoms.

For the second parent, the gen_iptm is 0.7640, the gen_plddt is 0.6997, and the gen_rmsd is 0.4960 Å. The gen_ligand_iptm is 0.7640, and the gen_min_interaction_pae is 1.7476. The gen_complex_iplddt is 0.6650, and the gen_interaction_pae is 5.7338. The gen_complex_pde is 0.8541, the gen_ptm is 0.7582, and the gen_target_ptm is 0.9534. The refolded delta SASA is 352.8596 Ų, with an affinity prediction value of 0.9100 and a binary probability of 0.4241.

The target-level ADMET aggregate risk score of 0.4776 and the QED of 0.5346 are ligand properties, not candidate properties. They describe pentachlorophenol itself. The ADMET toxicity risk score of 0.9625 is the standout value: the ligand is predicted to carry substantial toxicity risk, which is expected for a fully chlorinated phenol. The synthetic accessibility score of 0.8478 indicates that the ligand is easy to synthesize, which is useful for experimental follow-up.

Validation Plan

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

The first is a yeast surface display screen using IDT oPools oligo pools covering all 109 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This is the standard first-pass experimental filter for designed binders, and the tiering by generator confidence means the top parent and its proofread variants will be interrogated together as a cluster.

The second is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 20 candidates. The construct context includes the T7 promoter, GG, SD, Kozak, ATG, and FLAG tag, using the E. coli MFC codon table. This assay is designed to test binding in a cell-free format and is ready to run.

The order files and protocol documents are downloadable from the target's experiments tab.

Honest Assessment

This campaign produced a coherent set of ligand-conditioned binder designs for pentachlorophenol, with a clear top parent and a well-defined variant cluster. The pipeline generated 109 candidates, passed 20 through the generator prefilter with zero exclusions, and produced a best docking score of -6.69 kcal/mol across the campaign. The top-ranked parent by generator confidence, BG_SM_DENOVO_PCI_E2E_boltzgen_binder_009, docks at -4.36 kcal/mol with a generator confidence of 0.8803, a soluble classification, zero buried unsatisfied polars, and a SAP3D spatial score of -2.3692. Those are respectable numbers for a single-site design problem where the ligand is a rigid, heavily chlorinated aromatic with only one polar edge.

The main limitation is the docking score spread. The best campaign docking score of -6.69 kcal/mol is meaningfully better than the top parent's -4.36 kcal/mol, which suggests that the generator confidence ranking and the docking ranking are not perfectly aligned. A follow-up campaign should weight docking score more heavily in the final ranking, or run a second docking pass with a larger mode count to see whether the top parent's pose improves. The GNINA CNN pose score of 0.7144 for the top parent is moderate, and the affinity prediction value of 0.4744 is below the midpoint of the scale, so there is room to improve the predicted binding energetics.

The proofread variants are the other open question. Six of the eight listed rows are variants of two parents, and the variant rows carry their own source tool and structure source values but not the full developability metric set. The proofreading stage is clearly exploring local sequence space, but the campaign data surface does not tell us whether any variant improves on its parent's docking score or developability profile. A follow-up campaign should run the full docking and developability pipeline on the proofread variants individually, rather than relying on the parent's score to represent the cluster.

The ligand itself is a hard target. Pentachlorophenol's ADMET toxicity risk score of 0.9625 is a serious liability, and the QED of 0.5346 reflects the challenge of turning a fully chlorinated phenol into a drug-like molecule. The synthetic accessibility of 0.8478 is the one bright spot on the ligand side: the target is easy to obtain, which makes the planned YSD and IVTT experiments practical to run. The pharmacophore profile, with five hydrophobic features and one polar edge, means the designed binders must do most of their work through shape complementarity and hydrophobic burial. The top parent's delta SASA of 311.7009 Ų suggests that burial is happening, but the zero PLIP hydrogen bonds and zero salt bridges in the refolded structure indicate that the polar edge of the ligand is not being engaged as tightly as it could be. A follow-up campaign should explicitly reward designs that place a hydrogen-bond donor or acceptor near the phenolic oxygen, rather than relying on hydrophobic contacts alone.

The validation plan is ready to run, and the tiered YSD screen over all 109 candidates is the right first experiment. The IVTT TR-FRET assay on 20 candidates will provide a quantitative binding readout for the top tier. If those experiments confirm the ranking, the next design cycle should focus on improving the polar contact at the phenolic oxygen and pushing the docking score toward the campaign best of -6.69 kcal/mol.