TFA Binder Design: A Ligand-Conditioned Campaign Against Trifluoroacetic Acid

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

Trifluoroacetic acid (TFA) is a small, highly fluorinated carboxylic acid with the SMILES string O=C(O)C(F)(F)F and the CCD code TFA. At 114.02 Da, it is a compact ligand with a single carboxylic acid group and three fluorine atoms that dominate its electronic character. The molecule carries one hydrogen bond donor and one hydrogen bond acceptor, with zero rotatable bonds, making it a rigid, chemically defined target. Its cLogP of 0.63 places it near the polar edge of typical small molecule space, while its QED drug-likeness score of 0.5011 reflects the challenge of treating such a small, highly polar molecule as a conventional drug-like ligand.

TFA matters as a target because it represents a class of small, acidic metabolites and counterions that frequently appear in biochemical assays, industrial processes, and as a metabolic byproduct. Designing a protein binder that specifically recognizes TFA, rather than larger or more hydrophobic carboxylates, tests the ability of ligand-conditioned design to discriminate among closely related small molecules using only the ligand's shape, electrostatics, and hydrogen bonding features. The campaign described here used the AffiniBind small molecule pipeline to generate and rank 190 candidate protein binders against this target.

Ligand Preparation

The ligand preparation stage began with the TFA SMILES string and generated a three-dimensional conformer with partial charges assigned for downstream docking and scoring. Because TFA has zero rotatable bonds, the conformer generation step produced a single rigid geometry, which simplifies the pharmacophore fingerprinting and docking stages. The ligand's small size and high fluorine content mean that the partial charge assignment must accurately capture the strong electron-withdrawing effect of the trifluoromethyl group on the carboxylic acid, which influences both hydrogen bonding and electrostatic complementarity with any designed protein pocket.

The computed ligand properties establish the baseline for the campaign. TFA has a molecular weight of 114.02 Da, a cLogP of 0.63, one hydrogen bond donor, one hydrogen bond acceptor, and zero rotatable bonds. It shows zero Lipinski violations, which is expected for such a small molecule. The synthetic accessibility score of 0.8889 on the SAScore scale (where 1 is easy and 10 is hard) confirms that TFA itself is trivially accessible, though this metric describes the ligand, not the protein binders designed against it. The ADMET aggregate risk score of 0.3611 and the ADMET toxicity risk score of 0.7364 indicate that TFA carries a moderate toxicity liability, consistent with its known corrosive and irritant properties as a strong acid.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified four features on TFA: two single atom acceptors, one single atom donor, and one acidic group. The two acceptor features arise from the carbonyl oxygen and the fluorine atoms, which can act as weak hydrogen bond acceptors. The single donor feature corresponds to the carboxylic acid hydroxyl group. The acidic group feature captures the deprotonatable carboxylic acid, which at physiological pH will exist predominantly as the trifluoroacetate anion.

This pharmacophore profile is sparse but chemically coherent. A designed binder must satisfy the hydrogen bonding requirements of the carboxylic acid while also accommodating the electron-rich trifluoromethyl group. The acidic group feature is particularly important because it suggests that a complementary pocket should include a basic residue or a hydrogen bond donor capable of interacting with the deprotonated carboxylate. The absence of aromatic or hydrophobic pharmacophore features means that shape complementarity and electrostatic interactions, rather than pi-stacking or extensive hydrophobic burial, will dominate the binding interface.

Binder Generation

The binder generation stage produced 190 candidates using the AffiniBind pipeline version 0.5.0. The source tool split shows 80 candidates generated by boltzgen alone and 110 candidates generated by boltzgen+proofread, where the proofread step applies additional structural refinement to the initial BoltzGen output. All 190 candidates carried a generator confidence score, and 80 candidates passed the generator prefilter. The remaining 110 candidates joined the table after the prefilter stage as variant or union rows, meaning they were never screened by the prefilter but were carried forward for downstream assessment.

The per-tool metric summaries reveal a meaningful difference between the two generation routes. The 80 boltzgen candidates had a median generator confidence score of 0.6839, a median solubility score of 0.8180, a median SAP3D spatial score of -1.0827, a median average packing density of 15.8272, and a median flexible loop fraction of 0.4155. The 110 boltzgen+proofread candidates had a higher median generator confidence score of 0.7873.

Top Candidates

The ranked candidate table lists eight rows representing four distinct designs. Six of the rows are variants of two parent designs, meaning the ranking reflects the parent's score propagated to its variants. The two parent designs are BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_048 and BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_008, each with three proofread variants.

The table presents the two parent designs that carry the full set of developability metrics. The top-ranked parent, BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_048, achieved a generator confidence score of 0.8410 and a docking score of -3.61 kcal/mol. Its solubility score of 0.7267 places it in the soluble class, and its SAP3D spatial score of -0.8110 indicates a reasonably well-packed structure. The average packing density of 18.2308 and flexible loop fraction of 0.3333 describe a compact, moderately flexible binder.

The second parent, BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_008, scored 0.8322 on generator confidence and -2.57 kcal/mol on docking. Its solubility score of 0.6884 also places it in the soluble class, with a SAP3D spatial score of -0.7855, average packing density of 17.1467, and flexible loop fraction of 0.3600. Both parents show zero protease exposed motifs and zero buried unsatisfied polar counts, which are favorable developability signals.

The six variant rows, named with the __pf1_XX suffix, come from the boltzgen+proofread source tool and carry only their source tool, ligand gate status, and structure source. These variants represent proofread refinements of the two parent designs, and their presence in the ranked list reflects the parent's score propagated to its variants. The campaign generated more designs than the eight rows shown here, as the full campaign includes 190 candidates across both source tools.

Docking and Developability

The docking stage used AutoDock Vina with 10 modes per candidate, and the best mode was mode 1 for both parent designs. The top parent, BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_048, achieved a docking score of -3.61 kcal/mol with a GNINA CNN pose confidence of 0.9025 and a GNINA CNN affinity of 3.3020. The second parent, BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_008, docked at -2.57 kcal/mol with a GNINA CNN pose confidence of 0.7189 and a GNINA CNN affinity of 2.8660.

These docking scores are modest in absolute terms, which is expected for a small, highly polar ligand like TFA. The GNINA CNN pose confidence of 0.9025 for the top candidate suggests that the docking pose is geometrically plausible, even if the predicted affinity is not strong. The generator also produced structural quality metrics for the parent designs. The top parent had a gen_iptm of 0.7765, a gen_plddt of 0.7162, and a gen_rmsd of 0.4446 Å. Its gen_ligand_iptm of 0.7765 and gen_min_interaction_pae of 1.9884 indicate that the ligand is predicted to be well integrated into the designed interface. The gen_complex_iplddt of 0.5925 and gen_interaction_pae of 5.5210 describe a moderately confident complex prediction.

The second parent showed a gen_iptm of 0.7785, a gen_plddt of 0.6350, and a gen_rmsd of 0.5781 Å. Its gen_ligand_iptm of 0.7785 and gen_min_interaction_pae of 1.6680 are comparable to the top parent, while its gen_complex_iplddt of 0.5619 and gen_interaction_pae of 6.7771 suggest a slightly less confident complex prediction. Both parents showed zero raw chain breaks, zero unplaced atoms, and no chain break flags, indicating that the generated structures are chemically complete.

The developability metrics for the parent designs are encouraging. Both parents have solubility scores above 0.68 and are classified as soluble. Their SAP3D spatial scores of -0.8110 and -0.7855 are negative, which is the favorable direction for this metric. The average packing densities of 18.2308 and 17.1467 are within a reasonable range for de novo designed proteins. The flexible loop fractions of 0.3333 and 0.3600 indicate that about a third of each binder is in flexible loops, which is typical for small designed binders.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental validation routes. These experiments 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 190 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen would provide a high-throughput readout of binding to TFA in a display format.

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 with a GG + SD + Kozak + ATG + FLAG construct context and the E. coli MFC codon table. This assay would provide a quantitative binding readout in a cell-free format.

The order files and protocol documents are downloadable from the target's experiments tab. These planned experiments are ready to run and available for another lab to reproduce.

Honest Assessment

This campaign against trifluoroacetic acid produced a ranked set of 190 candidate binders with a clear top design, BG_SM_DENOVO_TFA_L80_E2E_boltzgen_binder_048, achieving a generator confidence score of 0.8410 and a docking score of -3.61 kcal/mol. The pipeline successfully carried the ligand through preparation, pharmacophore fingerprinting, single-site design, binder generation, docking, and ADMET assessment without losing candidates to the prefilter, since all 80 prefilter-screened candidates passed and the remaining 110 joined as proofread variants.

What worked well is the structural quality of the generated binders. Both parent designs show zero chain breaks, zero unplaced atoms, zero protease exposed motifs, and zero buried unsatisfied polar counts. Their solubility scores place them in the soluble class, and their SAP3D spatial scores are favorable. The generator's confidence in the ligand interface, reflected in gen_ligand_iptm values of 0.7765 and 0.7785, suggests that the designed pockets do engage the TFA pharmacophore features identified earlier: the two acceptors, one donor, and one acidic group.

What the pipeline struggled with is the absolute binding affinity. A docking score of -3.61 kcal/mol for the top candidate is weak for a designed binder, and the GNINA CNN affinity of 3.3020 reinforces that the predicted interaction is not strong. The second parent's docking score of -2.57 kcal/mol is even weaker. This is likely a consequence of TFA's small size and high polarity: there is simply not enough surface area or hydrophobic character to drive a high-affinity interaction, and the strong solvation of the trifluoroacetate anion in water imposes a large desolvation penalty on any binder.

The ADMET aggregate risk score of 0.3611 for TFA itself is moderate, but the toxicity risk score of 0.7364 is a reminder that TFA is a corrosive strong acid. Any binder designed against TFA would need to be robust to the acidic conditions under which TFA is typically handled, which the current developability metrics do not directly address.

A follow-up campaign would target a different aspect of the problem. The pharmacophore analysis identified only four features, all of them polar or ionic. A more productive approach might be to design binders against a TFA conjugate or a TFA-containing metabolite where additional hydrophobic or aromatic features are present, giving the generator more chemical handles to build a high-affinity pocket. The planned yeast surface display and IVTT TR-FRET experiments, with order files already available, provide a clear path to test whether the computational ranking holds up in the laboratory.