A ligand-conditioned design campaign against perfluorobutanesulfonamide generated 195 binder candidates, with a top docking score of -7.74 kcal/mol and a target QED of 0.806.
Explore more from this discipline or browse the full archive.
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.
Send us a target surface and we will scope a feasibility review, design binders, and report the same objective metrics our research is built on.
A full AffiniBind campaign against perfluorononanoic acid generated 194 binder candidates, with the top design reaching a docking score of -9.50 kcal/mol and a generator confidence of 0.9549.
A ligand-conditioned design campaign targeting perfluorooctanoic acid produced 187 candidate binders, with the top-ranked design reaching a docking score of -8.26 kcal/mol and a generator confidence of 0.902.
Perfluorobutanesulfonamide (PFBSA) is a four-carbon perfluoroalkyl sulfonamide with the SMILES string NS(=O)(=O)C(F)(F)C(F)(F)C(F)(F)C(F)(F)F. The molecule carries a terminal sulfonamide group attached to a fully fluorinated butyl chain, giving it a compact polar head and a hydrophobic, fluorinated tail. PFBSA belongs to the broader class of perfluoroalkyl substances that have drawn sustained attention in environmental and toxicological chemistry, and its sulfonamide head group distinguishes it from the corresponding sulfonic acid. The CCD code for this ligand is KPQ, and the campaign treated PFBSA as the binding interface for single-site protein binder design.
The ligand's computed properties place it in an interesting region of chemical space. Its molecular weight is 299.11 Da, with a cLogP of 1.70, one hydrogen-bond donor, two hydrogen-bond acceptors, and three rotatable bonds. It shows zero Lipinski violations. The QED drug-likeness score is 0.8063, which is a strong value for a molecule of this size and fluorination pattern. The synthetic accessibility score is 0.7833 on the standard SAScore scale where 1 is easy and 10 is hard, indicating a readily accessible compound. The ADMET aggregate risk score is 0.5068, while the ADMET toxicity risk score is 0.9006, a high value that reflects the known persistence and bioaccumulation concerns associated with perfluoroalkyl substances. These target-level properties frame the design problem: PFBSA is a synthetically tractable, drug-like small molecule with a significant toxicity liability, making it a relevant model target for testing whether a designed protein binder can recognize and engage a fluorinated sulfonamide.
The ligand preparation stage began with the PFBSA SMILES string and the CCD code KPQ. RDKit conformer generation produced a three-dimensional starting geometry for the ligand, and partial charges were assigned to prepare the molecule for downstream pharmacophore fingerprinting and docking. The ligand's conformational flexibility is modest, with three rotatable bonds, all of which lie along the perfluorobutyl chain. The sulfonamide head group is relatively rigid, with the nitrogen and two sulfonyl oxygens forming a fixed hydrogen-bonding surface.
The prepared ligand served as the single-site design interface. In this pipeline configuration, the ligand itself defines the binding site; there is no patch discovery step. The protein binder generation therefore had to build a complementary pocket directly around the PFBSA molecule, using the ligand's shape, charge distribution, and pharmacophore features as conditioning signals. The preparation stage's output fed directly into pharmacophore fingerprinting, where the ligand's chemical features were encoded for the generative models.
The pharmacophore fingerprinting stage identified three features on PFBSA. The feature counts are two Acceptor.SingleAtomAcceptor features and one Donor.SingleAtomDonor feature. These correspond to the sulfonamide group's two sulfonyl oxygens acting as hydrogen-bond acceptors and the sulfonamide nitrogen acting as a hydrogen-bond donor. The perfluorobutyl chain contributes no pharmacophore features in this encoding, consistent with its hydrophobic, nonpolar character.
This pharmacophore profile is compact but chemically informative. The three features cluster at the ligand's polar head, meaning the designed binder must satisfy a concentrated hydrogen-bonding demand at one end of the molecule while accommodating the fluorinated tail through shape complementarity and van der Waals contacts. The pharmacophore fingerprint, along with Morgan and MXFP fingerprints and GNN embeddings from molfeat, provided the conditioning input for the generative models. The single-site design constraint meant that the generated protein sequences had to place complementary side chains precisely at the sulfonamide head while building a pocket that wraps the fluorinated chain.
The binder generation stage produced 195 candidates in total. The source-tool split shows 80 candidates from boltzgen and 115 from boltzgen+proofread. The generator prefilter passed 80 candidates, and the prefilter excluded zero candidates. The remaining 115 candidates joined the table after the prefilter stage as variant or union rows, meaning they were never screened by the prefilter. These 115 candidates are the proofread variants of the 80 boltzgen-generated parent designs.
The generator confidence scores are available for all 195 candidates. The median generator confidence score for the 80 boltzgen candidates is 0.7062, while the median for the 115 boltzgen+proofread candidates is 0.8122. The proofread variants therefore carry a higher median confidence than the original boltzgen outputs, consistent with the proofreading step refining the generated sequences. The boltzgen candidates also carry median developability metrics: a median solubility score of 0.8136, a median SAP3D spatial score of -1.0113, a median average packing density of 15.6936, and a median flexible loop fraction of 0.4035.
The top-ranked candidates in the campaign table represent three distinct designs among the eight listed rows. Seven of the eight rows are variants of two parent designs. The parent design BG_SM_DENOVO_KPQ_L80_E2E_boltzgen_binder_041 appears with five proofread variants, and the parent design BG_SM_DENOVO_KPQ_L80_E2E_boltzgen_binder_030 appears with two proofread variants. The ranking reflects the parent's generator confidence score propagated to its variants, so the variant cluster from binder_041 occupies the top six positions, followed by the two variants from binder_030.
The top-ranked candidate is BG_SM_DENOVO_KPQ_L80_E2E_boltzgen_binder_041, a boltzgen-generated design with a generator confidence score of 0.8906. This parent design carries a full set of developability and docking metrics. Its solubility score is 0.8479, classified as soluble, and its SAP3D spatial score is -1.1218. The average packing density is 15.7808, the protease-exposed motif count is 1.0000, and the buried unsatisfied polar count is 0.0000. The flexible loop fraction is 0.3699. The generated structure metrics include an iPTM of 0.8415, a pLDDT of 0.7555, an RMSD of 0.6363, and a ligand iPTM of 0.8415. The docking score for this candidate is -3.85 kcal/mol, with a GNINA CNN pose score of 0.5673 and a CNN affinity value of 3.6850. The docking run used 10 modes, and the best mode was mode 1.
The five proofread variants of binder_041 (__pf1_02, __pf1_03, __pf1_04, __pf1_00, and __pf1_01) each carry a source tool label of boltzgen+proofread, a ligand gate status of not_run, and a structure source of refold. The two proofread variants of binder_030 (__pf1_04 and __pf1_03) carry the same source tool, ligand gate status, and structure source labels. These variant rows represent the proofread outputs derived from their respective parent designs.
| Candidate | Source Tool | Generator Confidence | Solubility Score | SAP3D Spatial Score | Average Packing Density | Flexible Loop Fraction |
|---|---|---|---|---|---|---|
| BG_SM_DENOVO_KPQ_L80_E2E_boltzgen_binder_041 | boltzgen | 0.8906 | 0.8479 | -1.1218 | 15.7808 | 0.3699 |
The table above presents the parent design binder_041, the only candidate among the top eight listed rows that carries the full developability metric set. The generator confidence score of 0.8906 is the highest in the listed set, and the solubility score of 0.8479 places this design in the soluble class. The SAP3D spatial score of -1.1218 indicates favorable spatial packing, and the average packing density of 15.7808 is consistent with a well-packed core. The flexible loop fraction of 0.3699 suggests a moderately flexible structure, which may be relevant for induced-fit recognition of the ligand.
The proofread variants of binder_041 and binder_030 are described by their source tool, ligand gate status, and structure source. All seven variant rows carry the boltzgen+proofread source tool label, a ligand gate status of not_run, and a structure source of refold. The proofread step produced these variants from the parent designs, and the campaign's median generator confidence score for the boltzgen+proofread group is 0.8122, higher than the boltzgen-only median of 0.7062.
The campaign's best docking score is -7.74 kcal/mol, achieved across the full candidate set. The top-ranked parent design binder_041 has a docking score of -3.85 kcal/mol, which is less favorable than the campaign best but still indicates a measurable binding pose. The GNINA CNN pose score for binder_041 is 0.5673, and the CNN affinity value is 3.6850. The docking run for this candidate used 10 modes, with the best mode identified as mode 1.
The developability metrics for the boltzgen candidates show a median solubility score of 0.8136, a median SAP3D spatial score of -1.0113, a median average packing density of 15.6936, and a median flexible loop fraction of 0.4035. These medians reflect the 80 boltzgen-scored candidates. The parent design binder_041 exceeds the median on solubility (0.8479 versus 0.8136), SAP3D spatial score (-1.1218 versus -1.0113), and average packing density (15.7808 versus 15.6936), while its flexible loop fraction of 0.3699 is below the median of 0.4035.
The target-level ADMET assessment applies to PFBSA itself, not to the protein binders. The ligand's ADMET aggregate risk score is 0.5068, and its ADMET toxicity risk score is 0.9006. The high toxicity risk score is a property of the perfluoroalkyl sulfonamide class and does not reflect the designed binders. The ligand's QED drug-likeness score of 0.8063 and synthetic accessibility score of 0.7833 are target-level properties that describe the small molecule's tractability as a chemical matter.
The campaign's ordering stage produced order files for two planned experiments. These experiments are available for reproduction and have not been run.
The first planned experiment is a yeast surface display (YSD) 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 high-throughput experimental readout of binding activity for the full candidate set.
The second planned experiment is an IVTT TR-FRET assay using 96-well IDT gBlock constructs covering 80 candidates. The constructs use the T7 promoter plus GG plus SD plus Kozak plus ATG plus FLAG construct context, with the E. coli MFC codon table. This assay would provide a quantitative binding measurement for a focused subset of the candidates.
The order files and protocol documents are downloadable from the target's experiments tab. These resources are ready to run and available for another lab to reproduce.
This campaign against PFBSA produced a substantial candidate set with a clear structure. The pipeline generated 195 candidates, with 80 boltzgen parent designs and 115 proofread variants. The generator prefilter passed all 80 boltzgen candidates, and the proofread variants joined the table after the prefilter stage. The median generator confidence score improved from 0.7062 for boltzgen to 0.8122 for boltzgen+proofread, indicating that the proofreading step added value in terms of model confidence.
The top-ranked parent design binder_041 achieved a generator confidence score of 0.8906, a solubility score of 0.8479, and a SAP3D spatial score of -1.1218. Its docking score of -3.85 kcal/mol is moderate, and the campaign's best docking score of -7.74 kcal/mol suggests that other candidates in the full set achieved more favorable binding poses. The target ligand's QED of 0.8063 and synthetic accessibility of 0.7833 make PFBSA a tractable small molecule for binder design, while its ADMET toxicity risk score of 0.9006 highlights the environmental and toxicological relevance of this target class.
The pipeline's main challenge is evident in the candidate table structure. The eight listed rows represent only three distinct designs, with seven rows being proofread variants of two parent designs. The variant rows carry source tool, ligand gate status, and structure source labels, but the developability and docking metrics are concentrated on the parent design. A follow-up campaign would benefit from generating a larger number of distinct parent designs rather than producing many proofread variants of a small number of parents. The proofread step clearly improves confidence scores, but the diversity of the underlying design space is the limiting factor for identifying high-affinity binders.
The planned YSD screen covering all 195 candidates and the IVTT TR-FRET assay covering 80 candidates provide a clear experimental path forward. The YSD screen's tiered-by-confidence design is well suited to the candidate distribution, and the TR-FRET assay's focused subset of 80 candidates aligns with the boltzgen parent count. A follow-up campaign that increases the number of distinct parent designs, perhaps by varying the generative conditioning or the single-site design constraints, would give the experimental screens a broader sequence space to interrogate. The PFBSA target itself, with its compact pharmacophore and fluorinated tail, remains a useful test case for ligand-conditioned binder design, and the current campaign provides a solid foundation for iterative improvement.