PFBSA: Ligand-Conditioned Binder Design for Perfluorobutanesulfonamide

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

Perfluorobutanesulfonamide (PFBSA) belongs to the perfluoroalkyl sulfonamide family, a class of compounds that has drawn sustained attention in environmental chemistry and toxicology. The molecule carries a fully fluorinated four-carbon chain attached to a sulfonamide head group, giving it the amphiphilic character typical of short-chain perfluoroalkyl substances. Its persistence in environmental compartments and its detection in biological samples make PFBSA a relevant target for engineered recognition, whether for biosensing, environmental monitoring, or selective sequestration applications.

The ligand presents a compact but chemically demanding surface for protein recognition. The perfluorinated tail offers a hydrophobic patch with unusual polarizability characteristics, while the sulfonamide group provides hydrogen-bond donor and acceptor functionality. A binder that can discriminate PFBSA from structurally related perfluoroalkyl sulfonates and sulfonamides would need to read out both the fluorinated chain length and the specific head group chemistry. This campaign set out to generate such binders using the AffiniBind ligand-conditioned design pipeline.

Ligand Preparation

The target ligand was prepared from its SMILES string, NS(=O)(=O)C(F)(F)C(F)(F)C(F)(F)C(F)(F)F, with the CCD code KPQ. RDKit conformer generation produced the three-dimensional starting geometry, and partial charges were assigned for downstream docking and interaction scoring. The ligand is small by protein-binding standards, with a molecular weight of 299.11 Da, and it carries one hydrogen-bond donor and two hydrogen-bond acceptors. Its cLogP of 1.70 reflects the balance between the polar sulfonamide and the hydrophobic perfluorinated tail.

The ligand's drug-likeness profile is favorable. It has zero Lipinski violations, a QED score of 0.8063, and a synthetic accessibility score of 0.7833 on the SAScore scale where 1 is easy and 10 is hard. These values matter less for the binder design itself than for the broader context: PFBSA is a tractable small molecule for assay development and for any downstream applications where the ligand, rather than the protein, must be handled at scale. The ADMET aggregate risk score of 0.5068 and the toxicity risk score of 0.9006 flag the compound's toxicological profile, which is consistent with the known concerns around perfluoroalkyl substances.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified three features on PFBSA: two single-atom acceptors and one single-atom donor. These features map directly onto the sulfonamide head group, where the two sulfonyl oxygens serve as hydrogen-bond acceptors and the sulfonamide nitrogen serves as a hydrogen-bond donor. The perfluorinated tail contributes no classical pharmacophore features, which is expected: carbon-fluorine bonds are weak hydrogen-bond acceptors at best and do not register in standard pharmacophore definitions.

This feature profile has a practical consequence for binder design. The polar recognition surface is concentrated in a small region of the ligand, while the fluorinated tail must be accommodated through shape complementarity and van der Waals contacts rather than directional hydrogen bonds. A successful binder therefore needs a pocket that combines a precise polar interface at the sulfonamide with a hydrophobic channel or cleft sized for a four-carbon perfluorinated chain. The single-site design strategy, where the ligand itself defines the binding interface, is well suited to this challenge because it forces the generated protein to build directly around the ligand's available interaction surface.

Binder Generation

The campaign generated 120 candidates using the AffiniBind pipeline version 0.5.0. The generator prefilter accepted 20 candidates and excluded none, while 100 additional candidates joined the table after the prefilter stage as variant or union rows. The source-tool split shows 20 candidates from boltzgen and 100 from boltzgen+proofread, indicating that the proofreading step expanded the candidate pool fivefold relative to the initial generator output.

The generator confidence scores were available for all 120 candidates. The median generator confidence score for the 20 boltzgen candidates was 0.6926, and the same median value of 0.6926 held for the 100 boltzgen+proofread candidates. This consistency suggests that the proofreading step preserved the confidence distribution of the parent designs while expanding the pool of variants available for downstream filtering.

The boltzgen candidates carried additional developability metrics. Across the 20 scored candidates from this source, the median solubility score was 0.8288, the median SAP3D spatial score was -1.0686, the median average packing density was 15.6628, and the median flexible loop fraction was 0.3870. These values indicate that the generator produced candidates with generally favorable solubility and packing characteristics, though the flexible loop fraction suggests a moderate degree of conformational flexibility in the designed backbones.

Top Candidates

The ranked candidate table lists eight rows representing four distinct designs. Six of the rows are variants of two parent designs: five variants derive from BG_SM_DENOVO_KPQ_E2E_boltzgen_binder_015, and one variant derives from BG_SM_DENOVO_KPQ_E2E_boltzgen_binder_001. The ranking reflects the parent design's generator confidence score propagated to its variants, so the top five rows all carry the same confidence score of 0.8374 from the parent design, and the next three rows carry 0.8206 from the second parent.

The table reveals an interesting inversion between confidence and docking score. The top-ranked design by generator confidence, binder_015 at 0.8374, docks at -6.10 kcal/mol. The second-ranked design, binder_001 at 0.8206, docks slightly better at -6.27 kcal/mol. The difference is modest, 0.17 kcal/mol, but it suggests that the generator's confidence ranking does not perfectly track docking performance. Both designs show strong interface metrics: binder_015 has a gen_iptm of 0.8077 and a gen_ligand_iptm of 0.8077, while binder_001 has a gen_iptm of 0.8000 and a gen_ligand_iptm of 0.8000. The gen_plddt values of 0.7085 and 0.7906 respectively indicate that binder_001 has higher predicted local structure confidence despite its lower generator confidence score.

The variant rows for each parent carry identical docking scores and generator metrics to their parent, which is expected given that the proofreading step preserves the core design while introducing sequence or structural refinements. The practical takeaway is that the campaign's top candidates cluster around two distinct design solutions, both of which achieve docking scores in the -6.1 to -6.3 kcal/mol range.

Docking and Developability

The best docking score across the campaign was -7.24 kcal/mol, which exceeds the scores of the top-ranked candidates shown in the table. This indicates that the full candidate pool contains designs with stronger predicted binding than the highest-confidence representatives. The docking scores for the listed top candidates, -6.10 and -6.27 kcal/mol, represent moderate predicted affinity for a ligand of this size. The GNINA CNN pose scores of 0.7052 for binder_015 and 0.5325 for binder_001 provide an additional confidence readout on the docked poses, with higher values indicating greater pose plausibility.

The developability metrics for the two parent designs are encouraging. binder_015 has a solubility score of 0.8005 and is classified as soluble, with a SAP3D spatial score of -1.1717 and an average packing density of 15.3766. binder_001 shows a solubility score of 0.8470, also classified as soluble, with a SAP3D spatial score of -1.0967 and an average packing density of 16.4865. Both designs have zero buried unsatisfied polar atoms, which is a strong indicator of well-packed cores. The protease-exposed motif counts are low: 2.0 for binder_015 and 1.0 for binder_001, suggesting limited vulnerability to proteolytic degradation. The flexible loop fractions of 0.3766 and 0.3108 respectively indicate that binder_001 has a somewhat more rigid backbone, which may contribute to its higher pLDDT.

The ligand-level ADMET profile applies to PFBSA itself rather than to the protein binders. The aggregate ADMET risk score of 0.5068 and the toxicity risk score of 0.9006 reflect the known toxicological concerns associated with perfluoroalkyl substances. These values do not affect the developability of the protein binders, but they are relevant for any application where the ligand-binder complex might be used in a biological context.

Validation Plan

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

The first is a yeast surface display screen using IDT oPools oligo pools covering all 120 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen is designed to test binding of the displayed proteins to PFBSA in a high-throughput format, with the confidence tiering allowing efficient allocation of screening resources.

The second is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 20 candidates. The constructs include the T7 promoter, GG, SD, Kozak, ATG, and FLAG construct context using the E. coli MFC codon table. This assay format enables cell-free expression and direct binding measurement in a 96-well plate, providing a quantitative readout of binding affinity for the top candidates.

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

Honest Assessment

This campaign produced a focused set of ligand-conditioned binder candidates for PFBSA, with two distinct parent designs dominating the top of the confidence-ranked list. The pipeline executed cleanly through ligand preparation, pharmacophore fingerprinting, binder generation, docking, and developability assessment. The pharmacophore analysis correctly identified the sulfonamide head group as the primary polar recognition surface, and the generated binders appear to have built complementary pockets around this feature while accommodating the perfluorinated tail through shape complementarity.

What worked well: the generator produced candidates with strong interface metrics, including gen_iptm values above 0.80 for both top designs and gen_ligand_iptm values that match the overall interface confidence. The developability profile of the parent designs is solid, with soluble classifications, negative SAP3D spatial scores indicating good spatial packing, and zero buried unsatisfied polar atoms. The docking scores in the -6.1 to -6.3 kcal/mol range for the top-ranked candidates, combined with the campaign-best score of -7.24 kcal/mol, suggest that the pool contains binders with meaningful predicted affinity for a ligand of this size.

The pipeline's main challenge is the tension between generator confidence and docking performance. The highest-confidence design, binder_015, docks at -6.10 kcal/mol, while the second-ranked design, binder_001, docks at -6.27 kcal/mol. The campaign-best docking score of -7.24 kcal/mol belongs to a candidate outside the top confidence tier. This suggests that the generator's confidence score, while useful for ranking, does not fully capture the determinants of docking performance. A follow-up campaign would benefit from a ranking strategy that weights docking score more heavily, or from a two-stage selection where the top confidence candidates are re-ranked by docking and interface metrics before experimental validation.

The variant structure of the candidate pool also deserves attention. The 100 proofread variants derive from a small number of parent designs, which means the effective diversity of the campaign is lower than the raw candidate count suggests. The two parent designs represented in the top table both achieve similar docking scores, indicating that the generator converged on a relatively narrow region of sequence space. A follow-up campaign could explore more diverse parent designs, perhaps by varying the generator's conditioning parameters or by introducing targeted mutations in the binding pocket to improve complementarity with the perfluorinated tail.

The planned validation experiments are well conceived. The yeast surface display screen with 120 candidates provides a high-throughput first pass, while the IVTT TR-FRET assay with 20 candidates offers a quantitative readout for the most promising designs. The tiering by confidence is a reasonable starting point, though the docking score inversion suggests that the experimental tiering might benefit from incorporating docking results alongside generator confidence. The availability of order files and protocols makes this campaign reproducible, which is essential for iterative refinement of the binder designs.

In summary, this campaign established a solid foundation for PFBSA binder development. The two parent designs represent viable starting points for experimental validation, with favorable developability metrics and moderate predicted binding affinity. The main opportunities for improvement lie in better integration of docking scores into the ranking, increased diversity of parent designs, and targeted optimization of the binding pocket to improve complementarity with the perfluorinated tail. The planned experiments will provide the critical data needed to determine whether the computational predictions translate into measurable binding in the laboratory.