Designing Protein Binders Against TFA: A Ligand-Conditioned 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

Trifluoroacetic acid (TFA) is a small, highly fluorinated carboxylic acid that presents an unusual challenge for protein binder design. With a molecular weight of 114.02 Da and a cLogP of 0.63, TFA is compact and polar enough to engage hydrogen-bonding networks, yet its three fluorine atoms create a distinctive electron-withdrawing environment that shapes how any complementary protein surface must recognize it. The molecule carries one hydrogen-bond donor and one hydrogen-bond acceptor, with zero rotatable bonds, meaning the ligand is conformationally rigid. That rigidity is a double-edged sword: it removes entropic uncertainty in pose prediction, but it also demands that the designed pocket match the ligand's fixed geometry with high precision.

TFA is not a drug-like molecule in the conventional sense. Its QED drug-likeness score is 0.5011, and its ADMET aggregate risk score is 0.3611, with a notably elevated toxicity risk score of 0.7364. These values frame the campaign's purpose: TFA is a metabolite and counter-ion relevant in analytical and biochemical workflows, and a designed binder against it could serve as a recognition element in sensors, purification resins, or diagnostic assays. The synthetic accessibility score of 0.8889 on the SAScore scale (1 = easy, 10 = hard) confirms that the ligand itself is trivial to obtain, which makes it a practical target for binder development even if its drug-likeness is modest.

Ligand Preparation

The ligand preparation stage began with the SMILES string O=C(O)C(F)(F)F, corresponding to the CCD code TFA. RDKit conformer generation produced the three-dimensional coordinates used throughout the campaign, and partial charges were assigned to support downstream docking and interaction scoring. Because TFA has zero rotatable bonds, the conformer ensemble is effectively a single rigid geometry, which simplifies the pharmacophore fingerprinting and docking stages. The ligand's single carboxylic acid group is the dominant chemical feature: it can act as a hydrogen-bond donor through the hydroxyl, a hydrogen-bond acceptor through the carbonyl oxygen, and an ionizable acidic group under physiological conditions.

The computed ligand properties anchor the campaign's expectations. The molecular weight of 114.02 Da places TFA well below the typical small-molecule drug range, and its single donor and single acceptor limit the number of directional interactions a binder can exploit. The cLogP of 0.63 indicates modest hydrophobicity, while the zero rotatable bonds mean the ligand will not pay a conformational entropy penalty upon binding. These properties were computed once for the target ligand and apply to every candidate in the campaign.

Pharmacophore Analysis

The pharmacophore fingerprinting stage identified four features for TFA: two single-atom acceptors, one single-atom donor, and one acidic group. The two acceptors correspond to the carbonyl oxygen and, in a broader sense, the electron-rich fluorine environment, though the feature typing assigns them as single-atom acceptors. The single donor is the hydroxyl proton of the carboxylic acid. The acidic group feature captures the ionizable carboxyl, which under many buffer conditions will exist as the deprotonated trifluoroacetate anion.

These four features define the interaction space that any designed binder must satisfy. A successful pocket needs to position a hydrogen-bond donor toward one of the ligand's acceptors, a hydrogen-bond acceptor toward the ligand's donor, and a positively charged or polar residue to engage the acidic group. The compactness of TFA means these features are clustered within a small volume, so the binder must achieve high shape complementarity while simultaneously satisfying the directional hydrogen-bonding requirements. The pharmacophore fingerprint was used to condition the generative models, ensuring that candidate pockets were biased toward surfaces that could present complementary chemical functionality.

Binder Generation

The campaign generated 115 candidates using the AffiniBind platform, pipeline version 0.5.0. The source-tool split shows 20 candidates from boltzgen and 95 from boltzgen+proofread. The proofread step is a post-generation refinement that refolds and validates the generated structures, and it contributed the majority of the candidate pool. Of the 115 candidates, 20 passed the generator prefilter, and 0 were excluded by that prefilter. The remaining 95 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 docking and assessment.

The generator confidence scores are available for all 115 candidates. The median generator confidence score for the 20 boltzgen candidates is 0.7172, while the median for the 95 boltzgen+proofread candidates is 0.7191. These medians are close, indicating that the proofread step did not systematically inflate or deflate confidence. The boltzgen candidates also carry median solubility scores of 0.8185, median SAP3D spatial scores of -1.0261, median average packing density of 15.8401, and median flexible loop fraction of 0.4333. These developability metrics provide a baseline for the generated binders' structural quality.

The single-site design strategy treated the ligand as the binding interface, with no patch discovery step. This means the generative models were conditioned directly on the TFA structure and its pharmacophore features, producing protein pockets de novo rather than grafting onto an existing scaffold. The result is a set of designs that are, in principle, fully complementary to TFA from the first residue.

Top Candidates

The ranked table below lists eight candidate rows from the campaign, representing three distinct designs. Seven of the eight rows are variants of two parent designs: BG_SM_DENOVO_TFA_E2E_boltzgen_binder_007 and BG_SM_DENOVO_TFA_E2E_boltzgen_binder_003. The ranking reflects the parent design's generator confidence score propagated to its variants, so the top six rows all share the same confidence score of 0.8367, and the last two share 0.8001.

The top-ranked parent design, BG_SM_DENOVO_TFA_E2E_boltzgen_binder_007, carries a generator confidence score of 0.8367 and a docking score of -3.62 kcal/mol. Its five proofread variants (__pf1_00 through __pf1_04) inherit the same confidence score and docking score, reflecting that the proofread step preserved the parent's binding pose and interaction profile. The second parent design, BG_SM_DENOVO_TFA_E2E_boltzgen_binder_003, has a lower confidence score of 0.8001 and a weaker docking score of -2.91 kcal/mol, but its generated pLDDT of 0.8362 is higher than the top parent's 0.7493. This trade-off between confidence, docking score, and structural confidence is worth noting: the top-ranked design docks better but has a lower pLDDT, while the second design has a higher pLDDT but docks less favorably.

The generated interaction metrics provide additional resolution. The top parent design has a gen_iptm of 0.7904, a gen_ligand_iptm of 0.7904, and a gen_min_interaction_pae of 1.8282. Its gen_complex_iplddt is 0.6672, and the gen_interaction_pae is 6.6428. The second parent design has a gen_iptm of 0.7366, a gen_min_interaction_pae of 1.5048, and a gen_complex_iplddt of 0.6791. The lower interaction PAE for the second design suggests tighter predicted interfacial confidence, even though its docking score is weaker. Both designs show one hydrogen bond in the refolded PLIP analysis and zero salt bridges, with substantial buried surface area upon binding: 202.15 Ų for the top parent and 174.63 Ų for the second parent.

Docking and Developability

Docking validation used AutoDock Vina with 10 modes per candidate, and the best mode was consistently mode 1. The top parent design and all five of its proofread variants achieved a docking score of -3.62 kcal/mol, which is the best docking score in the campaign. The GNINA CNN pose score for these candidates is 0.8640, and the GNINA CNN affinity is 3.190. The second parent design and its variant achieved a docking score of -2.91 kcal/mol, with a GNINA CNN pose score of 0.8319 and a GNINA CNN affinity of 2.997.

The docking scores are modest in absolute terms. A score of -3.62 kcal/mol corresponds to a predicted dissociation constant in the millimolar range, which is weak for a high-affinity binder. However, TFA is a small, highly polar ligand with limited hydrophobic surface area, so the achievable docking scores are inherently constrained. The campaign's best docking score of -3.63 kcal/mol (rounded from -3.62) reflects this fundamental challenge: small, rigid, polar ligands offer fewer favorable contacts than larger, more hydrophobic molecules.

The developability metrics for the top parent design are encouraging. The solubility score is 0.7460, placing it in the soluble class. The SAP3D spatial score is -0.6621, and the average packing density is 17.2368. The flexible loop fraction is 0.4474, indicating that nearly half the structure is in flexible loops, which may affect stability. The protease-exposed motif count is 0.0000, and the buried unsatisfied polar count is 0.0000, both favorable for developability. The generator also reports zero raw chain breaks, zero unplaced atoms, and a maximum consecutive C-N distance of 1.37 Å, all consistent with a well-formed structure.

The affinity prediction from the generator is 2.5716 for the top parent, with a binary affinity probability of 0.1770. This low probability suggests the generator itself is not confident that the designed binder will achieve tight binding, which aligns with the modest docking score. The second parent design has an affinity prediction of 2.3330 and a binary affinity probability of 0.2536, slightly higher but still below 0.5.

Validation Plan

The campaign's ordering stage produced order files for two planned experimental validation routes. These experiments are available and ready to run, but they have not been executed. The order files and protocol documents are downloadable from the experiments tab.

The first planned experiment is a yeast surface display (YSD) screen using IDT oPools oligo pools covering all 115 candidates, tiered by generator confidence score. The oligos are designed for gap-repair cloning into the pCTCON2 vector, which is a standard workflow for displaying designed binders on the yeast surface and selecting for binding to a labeled target. The tiering by confidence means the highest-confidence candidates will be screened first, allowing rapid triage of the most promising designs.

The second planned experiment 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 tag context, using the E. coli MFC codon table. This cell-free expression and assay format is designed to measure binding in solution without the need for protein purification, and the FLAG tag enables detection. The 20 candidates selected for this assay correspond to the 20 candidates that passed the generator prefilter.

Honest Assessment

This campaign against TFA produced a coherent set of candidates with internally consistent metrics, but the absolute binding predictions are modest. The best docking score of -3.63 kcal/mol and the generator's own affinity probability of 0.1770 both point to weak predicted binding. TFA is a difficult target: it is small, rigid, highly polar, and offers only four pharmacophore features for a binder to engage. The designed pockets achieve shape complementarity and present the right hydrogen-bonding functionality, as evidenced by the single hydrogen bond in the refolded PLIP analysis and the substantial buried surface area of 202.15 Ų for the top parent, but the predicted affinity remains in the weak range.

The pipeline's strength in this campaign is the consistency between the generator confidence scores and the docking scores. The top-ranked parent design has both the highest confidence (0.8367) and the best docking score (-3.62 kcal/mol), and the second parent design ranks lower on both axes. The proofread variants faithfully preserve the parent's metrics, which suggests the refinement step is stable and does not introduce spurious variation. The developability metrics are also favorable: the top parent is soluble, has zero buried unsatisfied polar atoms, zero protease-exposed motifs, and zero chain breaks.

What the campaign struggled with is the fundamental difficulty of designing high-affinity binders against a ligand as small and polar as TFA. The docking scores cluster in the -2.9 to -3.6 kcal/mol range, which is weak for most applications. The generator's affinity predictions are similarly cautious, with binary probabilities below 0.26 for both parent designs. A follow-up campaign would benefit from targeting a larger or more hydrophobic ligand, or from incorporating explicit electrostatic complementarity terms to better engage the trifluoroacetate anion's negative charge. The current designs rely on a single hydrogen bond and van der Waals contacts, but a stronger binder would need to exploit the acidic group more aggressively, perhaps through a designed arginine or lysine side chain positioned to form a salt bridge with the deprotonated carboxylate. The planned YSD and IVTT TR-FRET experiments will provide the first experimental test of whether these computational predictions translate into measurable binding, and the order files are available for any lab to reproduce the campaign's validation plan.