Ligand-Conditioned Binder Design Against DDT: A WeaveSeq AffiniBind 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

DDT (1,1,1-trichloro-2,2-bis(4-chlorophenyl)ethane) is a persistent organochlorine compound whose environmental and toxicological legacy makes it a compelling target for engineered recognition. The molecule presents two para-chlorophenyl rings joined through a trichloromethyl-substituted central carbon, giving it a compact but heavily halogenated architecture. Its SMILES string, Clc1ccc(C(c2ccc(Cl)cc2)C(Cl)(Cl)Cl)cc1, encodes a structure with two rotatable bonds and no hydrogen-bond donors or acceptors. The CCD code for this ligand is 6WT, and the PDB ID associated with the co-crystal is DDT.

The ligand's computed properties frame the design problem. Its molecular weight is 354.49 Da, and its cLogP reaches 6.50, reflecting the hydrophobic character imposed by four chlorine atoms and two aromatic rings. The molecule carries one Lipinski violation, driven by that high cLogP. Its QED drug-likeness score is 0.5403, and its synthetic accessibility score is 0.8789 on the SAScore scale where 1 is easy and 10 is hard. The ADMET aggregate risk score is 0.4611, with a notably higher ADMET toxicity risk score of 0.7890. These target-level values are the correct reference points for drug-likeness and ADMET assessment throughout this campaign.

Ligand Preparation

The AffiniBind small-molecule pipeline begins with ligand preparation, and for DDT the starting point is the SMILES string above. RDKit conformer generation and partial charge assignment produce the three-dimensional ligand geometry that conditions all downstream binder design. The prepared ligand carries two aromatic rings, two rotatable bonds, and zero hydrogen-bond donors or acceptors. That absence of polar functionality means the binding interface must be built almost entirely from shape complementarity, hydrophobic packing, and halogen-adjacent contacts rather than conventional hydrogen-bonding networks.

The ligand's molecular weight of 354.49 Da places it in a size range where a designed protein pocket can fully enclose the molecule. The cLogP of 6.50 and the single Lipinski violation are target-level facts that shape how the campaign interprets docking and developability outputs. The synthetic accessibility score of 0.8789 indicates that the ligand itself is straightforward to obtain, which matters for any downstream experimental validation that requires a physical sample of DDT.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified 18 features for DDT, distributed across three feature types. The dominant feature class is Hydrophobe.ThreeWayAttach, with 14 instances. This is consistent with the ligand's structure: the central carbon is a three-way attachment point connecting the two chlorophenyl rings and the trichloromethyl group, and the surrounding chlorinated aromatic system presents multiple hydrophobic attachment surfaces. The Aromatic.Arom6 feature type contributes 2 instances, corresponding to the two para-chlorophenyl rings. The LumpedHydrophobe.RH6_6 feature type adds 2 more instances, capturing the lumped hydrophobic character of the six-membered rings.

This pharmacophore profile tells the design engine what to recognize. With 14 of 18 features classified as three-way-attached hydrophobes, the ligand's binding surface is dominated by a single, densely chlorinated hydrophobic core. The two aromatic features provide planar recognition surfaces, but the absence of any hydrogen-bond donor or acceptor features means the designed binder cannot rely on polar specificity. The pharmacophore fingerprint therefore steers the generator toward pockets that maximize buried hydrophobic surface area and shape complementarity around the central trichloromethyl junction.

Binder Generation

The campaign generated 105 candidates using the AffiniBind pipeline version 0.5.0. The source-tool split is 20 candidates from boltzgen and 85 from boltzgen+proofread. All 105 candidates carry a generator confidence score. The generator prefilter passed 20 candidates and excluded 0, while 85 candidates joined the table after the prefilter stage as variant or union rows that were never screened by the prefilter. Those 85 rows correspond to the proofread variants of the boltzgen parents.

The single-site design strategy treats the ligand as the binding interface, with no patch discovery step. The generator conditions on the prepared DDT structure to produce a complementary protein pocket. The per-tool median generator confidence score for the 20 boltzgen candidates is 0.5878, while the 85 boltzgen+proofread candidates have a median of 0.6113. The proofread step therefore shifts the confidence distribution upward, though the top-ranked candidate in the final table comes from the boltzgen source.

The 8 listed candidate rows in the top-candidates table represent 4 distinct designs, with 6 rows being variants of 2 parent designs. The campaign as a whole produced more designs than the table shows, and the ranking reflects parent scores propagated to their variants. The two parent designs that anchor the table are BG_SM_DENOVO_6WT_E2E_boltzgen_binder_013 and BG_SM_DENOVO_6WT_E2E_boltzgen_binder_002. The first parent carries five proofread variants (__pf1_00 through __pf1_04), and the second carries one (__pf1_00).

Top Candidates

The ranked table below lists the candidates that carry generator confidence scores and docking scores on their own lines. The two parent designs appear with their full metric sets, while the proofread variants appear with their source-tool and structure-source values.

The top-ranked parent design, BG_SM_DENOVO_6WT_E2E_boltzgen_binder_013, carries a generator confidence score of 0.7470 and a docking score of -8.70 kcal/mol. Its solubility score is 0.7419 with a soluble classification, and its SAP3D spatial score is -1.6280. The average packing density is 15.0133, and the flexible loop fraction is 0.2933. The generator's internal metrics for this design include an interface predicted TM score (gen_iptm) of 0.7358, a predicted local distance difference test (gen_plddt) of 0.6156, and a root-mean-square deviation (gen_rmsd) of 1.1099 Å. The ligand iptm matches the interface iptm at 0.7358, and the minimum interaction predicted aligned error is 1.7067. The complex iplddt is 0.5675, and the interaction predicted aligned error is 8.1345. The complex predicted distance error is 1.0881, the overall predicted TM score is 0.6732, and the target predicted TM score is 0.8353. The protein iptm is 0.0000, reflecting the single-site design where the ligand is the interface. The refolded structure shows zero PLIP hydrogen bonds and zero salt bridges, with a delta SASA of 479.7201 Ų. The affinity prediction value is 0.9018 with a binary probability of 0.5325. The raw chain-break count is 0.0000, the maximum consecutive C-N-A value is 1.3700, and the unplaced atom count is 0.0000. The docking GNINA CNN pose score is 0.9013, and the CNN affinity is 6.9210. The docking run used 10 modes, with the best mode being mode 1.

The second parent design, BG_SM_DENOVO_6WT_E2E_boltzgen_binder_002, carries a generator confidence score of 0.7188 and a docking score of -6.51 kcal/mol. Its solubility score is 0.7533 with a soluble classification, and its SAP3D spatial score is -2.0647. The average packing density is 15.2308, and the flexible loop fraction is 0.2308. The generator's internal metrics include a gen_iptm of 0.6859, a gen_plddt of 0.6667, and a gen_rmsd of 0.8392 Å. The ligand iptm is 0.6859, the minimum interaction predicted aligned error is 1.6052, and the complex iplddt is 0.5878. The interaction predicted aligned error is 8.3464, the complex predicted distance error is 0.8872, and the overall predicted TM score is 0.7521. The target predicted TM score is 0.7872, and the protein iptm is 0.0000. The refolded structure shows zero PLIP hydrogen bonds and zero salt bridges, with a delta SASA of 399.1803 Ų. The affinity prediction value is 0.6929 with a binary probability of 0.6231. The raw chain-break count is 0.0000, the maximum consecutive C-N-A value is 1.3800, and the unplaced atom count is 0.0000. The docking GNINA CNN pose score is 0.8128, and the CNN affinity is 5.9460. The docking run used 10 modes, with the best mode being mode 1.

The six proofread variants of these two parents carry their own source-tool designation (boltzgen+proofread) and structure-source value (refold). The five variants of binder_013 are __pf1_00, __pf1_01, __pf1_02, __pf1_03, and __pf1_04. The single variant of binder_002 is __pf1_00. These variant rows represent the proofread outputs that joined the table after the prefilter stage.

Docking and Developability

The docking stage used AutoDock Vina with GNINA CNN scoring. The best docking score across the campaign is -8.70 kcal/mol, achieved by the top-ranked parent design BG_SM_DENOVO_6WT_E2E_boltzgen_binder_013. That design's GNINA CNN pose score is 0.9013, and its CNN affinity is 6.9210. The second parent design docks at -6.51 kcal/mol with a CNN pose score of 0.8128 and a CNN affinity of 5.9460. Both docking runs used 10 modes, and both selected mode 1 as the best mode.

The developability metrics for the two parent designs show distinct profiles. The top design has a solubility score of 0.7419 and a soluble classification, while the second design scores 0.7533 and is also classified as soluble. The SAP3D spatial scores are -1.6280 and -2.0647 respectively, with the more negative value for the second design indicating a different spatial packing profile. The average packing densities are 15.0133 and 15.2308, both in a range consistent with well-packed cores. The flexible loop fractions are 0.2933 and 0.2308. The protease-exposed motif count is 1.0000 for both designs, and the buried unsatisfied polar count is 1.0000 for the top design and 0.0000 for the second design.

The target-level ADMET assessment applies to DDT itself, not to the protein binders. The ADMET aggregate risk score is 0.4611, and the ADMET toxicity risk score is 0.7890. The QED drug-likeness is 0.5403, and the synthetic accessibility is 0.8789. These values describe the ligand as a small molecule, and they frame the challenge of designing a protein binder against a compound that is itself hydrophobic, persistent, and toxicologically significant.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental routes. Both 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 105 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen would provide the first experimental readout of binding activity across the full candidate set.

The second planned experiment 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. This assay would provide a quantitative binding signal for the top tier of candidates.

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 against DDT, with the top parent design docking at -8.70 kcal/mol and carrying a generator confidence score of 0.7470. The pipeline's pharmacophore analysis correctly identified the ligand's dominant feature: 14 of 18 features are three-way-attached hydrophobes, matching the densely chlorinated, nonpolar architecture of DDT. The generator responded by producing designs with substantial buried surface area on refolding, with delta SASA values of 479.72 Ų and 399.18 Ų for the two parent designs. The docking scores of -8.70 and -6.51 kcal/mol are consistent with a designed pocket that can engage the ligand's hydrophobic surface, and the GNINA CNN pose scores of 0.9013 and 0.8128 indicate that the docked poses align with the generator's intended binding modes.

The campaign's main limitation is the narrow spread of distinct designs in the top-ranked table. The 8 listed rows represent only 4 distinct designs, with 6 rows being proofread variants of 2 parents. The proofread step shifted the median generator confidence score from 0.5878 to 0.6113, but the variant rows themselves carry only source-tool and structure-source values in the data surface. The ranking therefore reflects the parent designs' scores propagated to their variants, and the variant cluster does not represent independent top candidates. A follow-up campaign would benefit from generating a larger number of distinct parent designs before applying the proofread step, so that the top of the ranking is not dominated by a single design family.

The target-level ADMET profile also shapes what this campaign can claim. DDT's QED of 0.5403 and ADMET toxicity risk of 0.7890 are properties of the ligand, not the binders. A protein binder that recognizes DDT does not inherit the ligand's toxicity, but the high cLogP of 6.50 and the single Lipinski violation mean that any downstream application involving the ligand itself must account for its poor drug-likeness. The synthetic accessibility of 0.8789 is favorable, so obtaining DDT for experimental validation is straightforward.

The planned validation experiments are ready to run and would provide the first experimental test of these designs. The yeast surface display screen covers all 105 candidates, and the IVTT TR-FRET assay covers 20 candidates. A follow-up campaign would target a broader set of distinct parent designs, with particular attention to designs that can improve on the top docking score of -8.70 kcal/mol while maintaining the solubility and packing quality seen in the current parents. The pipeline's single-site design strategy, with the ligand as the interface, worked well for this hydrophobic target, and the pharmacophore fingerprinting correctly steered the generator toward shape-complementary, hydrophobic pockets. The next iteration should focus on diversifying the parent design pool before proofreading, so that the final ranking reflects a wider range of binding solutions rather than a small cluster of related variants.