Tyrosinase

Kojic acid is the textbook reference inhibitor of tyrosinase (PDB 2Y9X), the copper-dependent enzyme that governs melanin synthesis and a prime target for skin-lightening and anti-browning agents. Plant polyphenols such as arbutin and quercetin are also widely reported as tyrosinase inhibitors. This note is a methods re-execution, not the reproduction of a specific published value: we dock these three known compounds against the enzyme's catalytic pocket and ask whether a standard docking protocol ranks them in line with their established activity.

The three ligands were docked into the di-copper active site of 2Y9X with AutoDock Vina (20 A box centred on the detected catalytic residues, exhaustiveness 64, three random seeds), rescored with the GNINA convolutional network, profiled for interactions (PLIP) and drug-likeness (Lipinski, Veber, QED), and passed through the ML/DTI module for a predicted pIC50. An aggregated consensus score (ADS) combines these signals. No self-docking control was available for this run, so pose geometry is reported as indicative rather than validated.

The raw Vina scores are close and modest (about -5.5 to -6.1 kcal/mol), as expected for small phenolics in a shallow metal-centred pocket; by binding energy alone, arbutin (-6.1) would edge out the reference. Once the signals are combined, the reference kojic acid rises to the top (consensus 44 versus 27 to 32; predicted pIC50 4.97 versus 3.7 to 4.1), carried by its high ligand efficiency despite its small size. All three pass Lipinski and Veber.

Ranking candidates on a single raw docking score is unreliable, especially when they differ in size: a small, efficient reference can look weaker on absolute binding energy yet be the most active. Consensus scoring, ligand efficiency and an orthogonal ML estimate recover the expected order. The absence of a self-docking control is a real reporting gap: the pose geometry should be verified before any firm structural claim. One caveat on the ML estimate: the model is trained on ChEMBL activity data and may already have seen these well-known compound-target pairs, so the predicted pIC50 should be read as a consistency check, not an independent, leak-free prediction.