A ligand-conditioned design campaign targeting perfluoroheptanoic acid generated 196 candidate binders, with the top design reaching a docking score of -7.51 kcal/mol and a generator confidence of 0.9234.
Explore more from this discipline or browse the full archive.
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.
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 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.
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.
Perfluoroheptanoic acid (PFHpA) belongs to the family of perfluoroalkyl carboxylic acids that have drawn sustained attention as environmental contaminants and as model compounds for studying fluorinated ligand recognition. The molecule presents a fully fluorinated seven-carbon chain terminating in a carboxylic acid head group, a combination that creates an unusual binding surface: a hydrophobic, lipophobic fluorocarbon tail and a single ionizable acid group. For a protein engineer, PFHpA is a demanding target because the fluorinated chain offers no hydrogen-bond donors and only the carboxylate oxygen atoms can participate in polar contacts. A binder that recognizes PFHpA must therefore derive much of its specificity from shape complementarity, van der Waals contacts with the fluorinated tail, and electrostatic or hydrogen-bonding interactions with the acid head group.
The campaign described here used the AffiniBind small molecule pipeline to design protein binders against PFHpA. The ligand was prepared from its SMILES string, fingerprinted, and used as the conditioning input for single-site binder generation. The resulting candidates were docked against the ligand and assessed with developability metrics. The target-level properties of PFHpA itself frame the design problem: a molecular weight of 364.06 Da, a cLogP of 3.81, one hydrogen-bond donor, one hydrogen-bond acceptor, five rotatable bonds, and zero Lipinski violations. The ligand's QED drug-likeness score is 0.7539, its synthetic accessibility score is 0.8144 on the SAScore scale where 1 is easy and 10 is hard, and its ADMET aggregate risk score is 0.3805 with a toxicity risk component of 0.7186.
PFHpA entered the pipeline as 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, corresponding to the CCD code 4EI. The preparation stage generated conformers with RDKit and assigned partial charges, producing the three-dimensional ligand representation used for pharmacophore fingerprinting and subsequent docking. The ligand's conformational flexibility is limited to five rotatable bonds along the perfluorinated chain, but the fluorinated backbone is relatively rigid in practice, which helps constrain the search space for both fingerprinting and binder generation.
The computed ligand properties place PFHpA in an interesting region of chemical space. The cLogP of 3.81 reflects the hydrophobic character of the perfluorinated tail, while the single carboxylic acid group keeps the molecule formally compliant with Lipinski's rule of five. The QED score of 0.7539 indicates a moderately drug-like molecule, though the ADMET toxicity risk of 0.7186 signals that the fluorinated chain carries a substantial toxicological liability. These target-level properties are the correct reference points for the campaign: the binder design problem is not to improve PFHpA itself, but to recognize it with sufficient affinity and specificity.
The pharmacophore fingerprinting stage identified four features on PFHpA. The feature breakdown is:
The two single-atom acceptor features correspond to the oxygen atoms of the carboxylic acid group, which can accept hydrogen bonds from a designed binder. The single donor feature is the acidic proton of the carboxylic acid, available for donation to a complementary acceptor on the protein. The acidic group feature captures the ionizable character of the carboxylate, which at physiological pH will exist predominantly in its deprotonated, negatively charged form. This pharmacophore profile is sparse but informative: the entire polar recognition surface of PFHpA is concentrated in the carboxylic acid head group, while the perfluorinated tail contributes no pharmacophore features and must be recognized through shape and van der Waals complementarity alone.
The campaign generated 196 candidates using the AffiniBind pipeline version 0.5.0. The generation stage split into two source tools: 80 candidates came directly from boltzgen, and 116 candidates came from boltzgen with proofreading applied. All 196 candidates carried a generator confidence score. The generator prefilter passed 80 candidates and excluded 0, meaning that the 116 proofread candidates joined the table after the prefilter stage as variant or union rows that were never screened by the prefilter.
The per-tool median summaries reveal a meaningful difference between the two generation routes. The 80 boltzgen candidates had a median generator confidence score of 0.7302, a median solubility score of 0.7921, a median SAP3D spatial score of -1.1694, a median average packing density of 15.4530, and a median flexible loop fraction of 0.3904. The 116 boltzgen+proofread candidates had a median generator confidence score of 0.8736. The proofread candidates therefore carried a higher median confidence, consistent with the proofreading step selecting or refining designs that the generator scored more favorably.
The ranked candidate table lists eight rows representing three distinct designs. Seven of the eight rows are variants of two parent designs: six variants descend from BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041, and two variants descend from BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_065. The ranking reflects the parent design's generator confidence score propagated to its variants, so the variant cluster around binder_041 occupies the top six positions.
The parent design BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041 carries the richest set of measured values. Its generator confidence score is 0.9234, the highest in the listed set. The design has a solubility score of 0.8147 with a soluble classification, a SAP3D spatial score of -2.0859, an average packing density of 12.9412, one protease-exposed motif, zero buried unsatisfied polar atoms, and a flexible loop fraction of 0.3137. The generator's structural confidence metrics for this parent include an iptm of 0.8756, a pLDDT of 0.8350, an RMSD of 0.6527 Å, a ligand iptm of 0.8756, a minimum interaction PAE of 0.8638, a complex ipLDDT of 0.7976, an interaction PAE of 5.1918, a complex PDE of 0.4566, a PTM of 0.8628, and a target PTM of 0.8667. The PLIP analysis after refolding identified zero hydrogen bonds and zero salt bridges, with a delta SASA of 307.2033 Ų. The affinity prediction value is 1.2602 with a binary probability of 0.1360. The design has zero raw chain breaks, a maximum consecutive C-N-A value of 1.3900, and zero unplaced atoms.
The docking result for this parent design is a score of -7.51 kcal/mol, with a GNINA CNN pose confidence of 0.4475 and a CNN affinity of 4.0720. The docking run used 10 modes and selected mode 1 as the best.
| Candidate | Source Tool | Generator Confidence | Docking Score | Solubility Score | SAP3D Spatial Score | Average Packing Density |
|---|---|---|---|---|---|---|
| BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041 | boltzgen | 0.9234 | -7.51 kcal/mol | 0.8147 | -2.0859 | 12.9412 |
The table above presents the parent design BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041 with the full set of metrics that its own candidate line carries. The six proofread variants of this parent, named __pf1_00 through __pf1_04, each carry only their source tool, ligand gate status, and structure source on their own lines. The two variants of the second parent, BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_065__pf1_02 and __pf1_04, likewise carry their own source tool and structure source values. The campaign's best docking score of -9.01 kcal/mol appears in the campaign summary, and the best QED score of 0.754 and best ADMET risk of 0.381 are target-level values from the ligand properties block.
The docking stage used AutoDock Vina with GNINA CNN rescoring. The parent design BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041 achieved a docking score of -7.51 kcal/mol, which places it in a favorable range for a designed binder against a small, largely hydrophobic ligand. The GNINA CNN pose confidence of 0.4475 indicates moderate confidence in the predicted binding pose, while the CNN affinity value of 4.0720 reflects the rescored affinity estimate. The docking run explored 10 modes and selected mode 1 as the best-scoring pose.
The developability metrics for the parent design are encouraging. The solubility score of 0.8147 with a soluble classification suggests the design expresses in a soluble form. The SAP3D spatial score of -2.0859 is negative, which in the SAP3D convention indicates fewer spatial aggregation-prone regions. The average packing density of 12.9412 falls in a reasonable range for a folded protein. The flexible loop fraction of 0.3137 indicates that roughly one-third of the design's residues are in flexible loops, which is typical for a de novo binder. The design carries one protease-exposed motif and zero buried unsatisfied polar atoms, both favorable for stability and expression.
The generator's structural confidence metrics reinforce the docking result. The iptm of 0.8756 and pLDDT of 0.8350 indicate high confidence in the overall fold and interface. The ligand iptm of 0.8756 matches the overall iptm, suggesting the ligand is well integrated into the designed interface. The minimum interaction PAE of 0.8638 and interaction PAE of 5.1918 indicate that the ligand is positioned with low predicted alignment error relative to the protein. The complex ipLDDT of 0.7976 and complex PDE of 0.4566 further support the quality of the designed complex. The refolded structure shows a delta SASA of 307.2033 Ų, indicating substantial burial of surface area upon complex formation, consistent with a specific binding interaction.
The campaign's ordering stage produced order files for two planned experimental routes. The first is a yeast surface display screen using IDT oPools oligo pools covering all 196 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. The second is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 80 candidates, with the T7 promoter plus GG plus SD plus Kozak plus ATG plus FLAG construct context using the E. coli MFC codon table. Both experiments are planned and available for reproduction; they have not been run. The order files and protocol documents are downloadable from the experiments tab.
This campaign produced a focused set of designed binders against PFHpA, with the parent design BG_SM_DENOVO_4EI_L80_E2E_boltzgen_binder_041 standing out on generator confidence (0.9234), docking score (-7.51 kcal/mol), and developability metrics including a soluble classification and a favorable SAP3D spatial score of -2.0859. The pharmacophore analysis correctly identified the carboxylic acid head group as the sole polar recognition element, and the generator responded by producing a design with zero buried unsatisfied polar atoms and a substantial delta SASA of 307.2033 Ų upon refolding. The proofread variants carried a higher median generator confidence (0.8736) than the direct boltzgen candidates (0.7302), suggesting the proofreading step added value in confidence ranking, though the variant lines themselves carry only source tool and structure source values, so the parent's docking and developability metrics remain the primary quantitative anchor for the cluster.
The pipeline's main challenge with PFHpA is the ligand's extreme polarity imbalance. The perfluorinated tail offers no pharmacophore features, so the generator had to rely on shape complementarity and van der Waals contacts for the majority of the ligand surface. The docking score of -7.51 kcal/mol for the top parent is respectable but not exceptional, and the GNINA CNN pose confidence of 0.4475 leaves room for improvement in pose prediction. The affinity prediction probability of 0.1360 is low, suggesting the generator's own affinity model was not strongly convinced of tight binding despite the favorable docking score. A follow-up campaign would benefit from enriching the training or conditioning signal for fluorinated ligands, perhaps by including explicit fluorophilic side-chain preferences or by weighting the pharmacophore fingerprint to emphasize the acid head group while allowing the tail to be recognized through shape alone. The planned YSD and IVTT TR-FRET experiments will provide the first experimental read on whether the computational confidence and docking scores translate into measurable binding.