A ligand-conditioned protein binder campaign targeting atrazine generated 118 candidates, with docking, developability, and ADMET metrics guiding the ranking of a compact set of parent designs and proofread variants.
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.
Atrazine is a chlorinated triazine herbicide with a compact, planar core and two secondary amine substituents. Its structure, CCNc1nc(Cl)nc(NC(C)C)n1, places a chlorine atom on the triazine ring, flanked by an ethylamino group and an isopropylamino group. That arrangement gives the molecule a molecular weight of 215.69 Da, a computed cLogP of 1.78, two hydrogen bond donors, five hydrogen bond acceptors, and four rotatable bonds. The ligand carries zero Lipinski violations and a QED drug-likeness score of 0.8019, which places it in a favorable range for small molecule recognition. Its synthetic accessibility score is 0.8400 on the SAScore scale where 1 is easy and 10 is hard, indicating a readily accessible target for binder campaigns.
The ADMET aggregate risk score for atrazine is 0.2938, with a toxicity risk score of 0.8098. That toxicity component is the dominant ADMET concern for this ligand, and it shapes how a designed binder might be used: the target is chemically tractable and synthetically simple, but its toxicological profile means that binder engineering efforts must be paired with careful downstream validation. The ligand is present in the Protein Data Bank under the CCD code ATZ, and the campaign treated it as the binding interface for single-site design, meaning the protein binder was generated directly against the ligand structure rather than discovered through patch mining.
The campaign began with RDKit conformer generation and partial charge assignment for atrazine. The SMILES string CCNc1nc(Cl)nc(NC(C)C)n1 was converted into a three-dimensional conformer set, and partial charges were assigned to support downstream docking and interaction scoring. The ligand's four rotatable bonds allow the ethylamino and isopropylamino groups to sample multiple conformations, but the triazine ring itself remains rigid, which constrains the pharmacophore geometry that a binder must complement.
The prepared ligand carried the CCD code ATZ, and its computed properties were recorded once at the ligand level. The molecular weight of 215.69 Da, cLogP of 1.78, and QED of 0.8019 all reflect the prepared structure. The synthetic accessibility score of 0.8400 and the ADMET aggregate risk score of 0.2938 were also computed at this stage, providing target-level context for the campaign's filtering and assessment steps.
The pharmacophore fingerprinting stage identified 10 features for atrazine. The feature breakdown includes three single-atom acceptors, one aromatic ring of type Arom6, two single-atom donors, three three-way-attach hydrophobes, and one lumped hydrophobe of type iPropyl. These features map directly onto the ligand's chemistry: the triazine ring contributes the aromatic feature, the ring nitrogens and the chlorine contribute acceptor character, the two secondary amines contribute donor features, and the isopropyl and ethyl substituents contribute hydrophobic contacts.
This pharmacophore profile guided the single-site design stage. The ligand was treated as the binding interface, and the generator was conditioned on the ligand structure to produce a complementary protein pocket. The presence of three acceptor features and two donor features means that a successful binder must satisfy a dense hydrogen-bonding network, while the three-way-attach hydrophobes and the iPropyl lumped hydrophobe require shape-complementary nonpolar contacts. The aromatic Arom6 feature anchors the planar triazine ring, which is the most rigid and geometrically demanding part of the ligand.
The campaign generated 118 candidates using the AffiniBind small molecule pipeline, version 0.5.0. The source-tool split shows 20 candidates from boltzgen and 98 from boltzgen+proofread. The generator prefilter passed 20 candidates and excluded 0, while 98 candidates joined the table after the prefilter stage as variant or union rows that were never screened by the prefilter. All 118 candidates carry a generator confidence score.
The 20 boltzgen candidates represent the primary generated designs, and the 98 boltzgen+proofread candidates are proofread variants of those parent designs. The per-tool median generator confidence score for boltzgen is 0.6264 across 20 scored candidates, and for boltzgen+proofread it is 0.6269 across 98 scored candidates. The boltzgen candidates also carry median solubility score of 0.8038, median SAP3D spatial score of -1.2020, median average packing density of 15.5744, and median flexible loop fraction of 0.3862.
The top-ranked candidate by generator confidence score is BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016, with a confidence score of 0.7538. This parent design carries a full set of developability and docking metrics. Its solubility score is 0.6411 with a soluble classification, its SAP3D spatial score is -1.6714, its average packing density is 14.5352, and its flexible loop fraction is 0.2394. The design has zero protease-exposed motif counts and zero buried unsatisfied polar counts, which are favorable developability signals.
The generator also produced interaction metrics for this parent design. The gen_iptm is 0.7059, gen_plddt is 0.7803, gen_rmsd is 0.6511, and gen_ipsae_min is 0.0940. The gen_ligand_iptm is 0.7059, gen_min_interaction_pae is 2.3392, gen_complex_iplddt is 0.6290, gen_interaction_pae is 8.9008, gen_complex_pde is 0.7474, gen_ptm is 0.7635, and gen_target_ptm is 0.8391. The gen_affinity_pred_value is 0.4879 with a binary probability of 0.2537. The structure source for this parent design is refold, and the ligand gate status is not_run.
The proofread variants of this parent design are BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016__pf1_01, __pf1_00, __pf1_04, __pf1_03, and __pf1_02. Each of these variants carries the parent field pointing to BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016, the source tool boltzgen+proofread, the ligand gate status not_run, and the structure source refold. A second parent design, BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_001, contributes two proofread variants: __pf1_02 and __pf1_01. These variants also carry the source tool boltzgen+proofread, ligand gate status not_run, and structure source refold.
The ranked candidate table lists eight rows representing three distinct designs. Seven of those rows are variants of two parent designs. The ranking reflects the parent design's generator confidence score propagated to its variants, so the variant cluster from BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016 occupies the top five positions after the parent itself, followed by the two variants of BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_001.
| Candidate | Source Tool | Generator Confidence | Solubility Score | SAP3D Spatial Score | Average Packing Density | Flexible Loop Fraction |
|---|---|---|---|---|---|---|
| BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016 | boltzgen | 0.7538 | 0.6411 | -1.6714 | 14.5352 | 0.2394 |
The parent design BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016 carries the full developability profile shown in the table. Its generator confidence score of 0.7538 is the highest in the listed set, and its solubility score of 0.6411 places it in the soluble class. The SAP3D spatial score of -1.6714 and average packing density of 14.5352 indicate a reasonably well-packed core, while the flexible loop fraction of 0.2394 suggests a moderately structured scaffold with some conformational flexibility.
The proofread variants of this parent design, and the variants of BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_001, carry their own source tool designation of boltzgen+proofread, their own ligand gate status of not_run, and their own structure source of refold. These variants represent the proofread outputs that were generated after the initial boltzgen designs, and they joined the campaign table after the prefilter stage.
The best docking score across the campaign is -5.43 kcal/mol. For the top-ranked parent design BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016, the docking score is -3.36 kcal/mol, with a docking_kcal_mol value of -3.36. The docking run produced 10 modes, with the best mode identified as mode 1. The GNINA CNN pose score for this candidate is 0.7581, and the GNINA CNN affinity value is 4.6570.
The docking score of -3.36 kcal/mol for the top parent design is moderate, and it sits below the campaign's best docking score of -5.43 kcal/mol. The GNINA CNN pose score of 0.7581 indicates that the docked pose has reasonable confidence, while the CNN affinity value of 4.6570 reflects the predicted binding affinity from the convolutional neural network component of the docking pipeline.
The developability metrics for the top parent design provide additional context. The solubility score of 0.6411 with a soluble classification, the SAP3D spatial score of -1.6714, and the average packing density of 14.5352 all point to a design that is developable in principle. The zero protease-exposed motif count and zero buried unsatisfied polar count are particularly favorable, as they reduce the risk of proteolytic susceptibility and energetic frustration in the core.
The generator-level interaction metrics for the top parent design include gen_iptm of 0.7059, gen_plddt of 0.7803, and gen_complex_iplddt of 0.6290. The gen_ligand_iptm of 0.7059 and gen_min_interaction_pae of 2.3392 indicate that the ligand is predicted to be well-integrated into the designed interface. The gen_interaction_pae of 8.9008 and gen_complex_pde of 0.7474 provide additional confidence in the predicted binding mode.
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 screen using IDT oPools oligo pools covering all 118 candidates, tiered by confidence, for gap-repair cloning into the pCTCON2 vector. This screen is designed to test binding of the generated candidates against atrazine in a high-throughput format.
The second planned experiment is an IVTT TR-FRET assay using 96-well IDT gBlock constructs for 20 candidates. The construct context includes the T7 promoter, GG, SD, Kozak, ATG, and FLAG tag, using the E. coli MFC codon table. This assay is designed to measure binding in a cell-free format with a smaller candidate set.
The order files and protocol documents are downloadable from the target's experiments tab. These planned experiments provide a clear path from the computational campaign to experimental validation, and they are ready to run for any lab that wants to reproduce or extend the campaign.
This campaign produced a compact set of parent designs and proofread variants targeting atrazine, with the top parent design BG_SM_DENOVO_ATZ_E2E_boltzgen_binder_016 carrying a generator confidence score of 0.7538, a soluble classification, and a docking score of -3.36 kcal/mol. The campaign's best docking score of -5.43 kcal/mol belongs to a different candidate in the full 118-candidate set, which means the ranking by generator confidence does not perfectly align with docking score. The ligand itself is chemically tractable, with a QED of 0.8019, a synthetic accessibility score of 0.8400, and an ADMET aggregate risk score of 0.2938, but its toxicity risk score of 0.8098 is a significant concern that any downstream application must address.
The pipeline's strengths in this campaign are the clean ligand preparation, the informative pharmacophore fingerprint with 10 features, and the generation of 118 candidates with a clear source-tool split. The proofread variants expanded the candidate pool substantially, with 98 boltzgen+proofread candidates joining the 20 boltzgen parents. The developability metrics for the top parent design are encouraging: zero protease-exposed motifs, zero buried unsatisfied polar counts, and a soluble classification all point to a design that could be expressed and tested.
The main challenge is the moderate docking score for the top-ranked parent design. A docking score of -3.36 kcal/mol is weaker than the campaign's best of -5.43 kcal/mol, and the GNINA CNN affinity value of 4.6570 suggests that the predicted binding affinity is modest. A follow-up campaign would benefit from re-ranking candidates by docking score rather than generator confidence alone, and from exploring whether the proofread variants of the top parent design can improve the docking score through additional refinement. The planned yeast surface display screen and IVTT TR-FRET assay will provide the experimental data needed to determine whether the computational ranking translates into measurable binding, and those experiments are available for another lab to reproduce using the order files and protocols on the experiments tab.