Supplementary information Molecules repartition into the various groups • 296 ligands with documented Ki from ChEMBL, from which: ◦ 232 ligands assigned to the training group (80%) ◦ 64 ligands assigned to the testing group (20 %) • For all docked ligands: ◦ 3 poses produced • Drugbank: 6631 molecules downloaded ◦ 148 could not be parsed by RDKit ◦ 232 molecules that could not be docked by Vina (eg: all molecules with inorganic atoms) ◦ 5078 non-FDA approved molecules, not used (eg: experimental drugs, illicit drugs) ◦ 291 FDA approved nutraceuticals, not used (eg: vitamin C, tetrahydrofolate) ◦ 1114 FDA approved drugs, used in the analysis. Redocking of the co-crystallised ligands of the PDB structures CES1 structure 1MX5 1MX9 1YA4 1YA8 1YAH Ligand Homatropine Naloxone Tamoxifen Mevastatin Ethyl acetate Ligand 2D structure Redocking Yes Yes Yes No, cleavage Yes applicable? products RMSD between 1.5 4.5 7.2 N/A 4.7 crystalized ligands and the best docked pose (Angstrom range on the best poses) Score of best pose -8.4 -7.9 -8.2 -8.0 -3.6 (~estimated Gibbs energy of binding in kcal/mol) Structure 1YAJ 2DQY 2DQZ 2DR0 2H7C Ligand Benzil Cholate Homatropine Taurocholate Coenzyme A Palmitate Coenzyme A Palmitate Ligand 2D structure Redocking applicable? Yes No Yes Yes Yes (outside active (Palmitate) site) RMSD between 3.5 N/A 3.7 6.5 8.7 crystalized ligands and the best docked pose (Angstrom range on the best poses) Vina score of best pose -5.2 -8.0 -5.3 -9.2 -7.2 (~estimated Gibbs energy of binding in kcal/mol) Figure S1: Tabulated results of the redocking of the co-crystallised ligand of each PDB structure, done on Autodock Vina with the same settings as in the main matter. Active site characterization according to fpocket 3.0 Default settings except for clustering distance (2.8 angstrom), leading to slightly overestimated volume and SASA, but yielding one and only one detected pocket covering the active site, allowing for structure-to-structure comparisons. (Addition of the value of multiple pocket is non-ideal for comparison purpose) Structure 1YAH 1YAJ 1YA8 2DQZ 1YA4 2H7C 1MX5 2DR0 2DQY 1MX9 Volume 701 795 950 958 984 1015 1071 1078 1157 1375 (active site, A³) Total SASA 185 189 227 224 222 274 276 258 269 356 Polar SASA 43 34 59 60 72 98 85 71 71 103 Apolar SASA 142 155 168 164 150 176 191 187 198 253 Figure S2: Active site volume and SASA for each structure, according to fpocket Complete dataset Complete dataset except ligands with Ki between 5 and 15 µM Threshold 10 µM 10 µM Number of active ligands 135 (50%) 140 (47%) Vina score only (no statistical model) Ensemble AUC (complete set) 0.55 0.57 Best AUC among all individual structures 0.60 0.62 Ensemble MLP model AUC (std. dev.) 5CV (training set) 0.83 (0.02) 0.81 (0.03) Ensemble LDA model AUC (std. dev.) 5CV (training set) 0.77 (0.07) 0.77 (0.08) Ensemble QDA model AUC (std. dev.) 5CV (training set) 0.77 (0.06) 0.77 (0.08) Figure S3: Comparison of the MLP, QDA and LDA models removing ligands with an activity between 5 µM and 15 µM. Table S1: 294 compounds with known activity on CES1 gathered in ChEMBL database. All_Docked_Molecules For each chemical, the ChEMBL ID, the SMILES code, the bioactivity (Ki) and the energy scores from autodock vina are depicted.