Browser-based molecular viewers, put together as vibe-coding experiments. Each page is self-contained – nothing is uploaded, everything runs locally in your browser.
A 96-compound semi-pure plate split into its six congeneric series, each as an R-group table where the depiction is oriented so only the substituent moves between rows. Potency is the purity-corrected pEC50, with the compounds whose correction rests on a low CAD recovery flagged. Two series are complete 12×2 matrices. A second tab overlays each series on its crystallized lead: every analogue was given a conformer ensemble, superimposed on the shared scaffold, and scored on shape plus feature overlap. Toggle ligands, show binding-site residues and hydrogen bonds, hover a residue for its id, and hover a ligand to draw its 2D structure below, registered to the series core.
One row per PXR cluster: the maximum common substructure drawn on the cluster representative, the distribution of pairwise RMSD over those MCS atoms, the distance matrix split into binding modes, a per-structure barcode of the protein–ligand interactions, and the per-residue RMSF of the binding site. Poses are compared in a common frame, so the RMSD measures where the shared core sits rather than its shape. Click a row to load that cluster's complexes into the 3D panel.
3D binding-site view of a PXR ligand series. Toggle ligands to overlay their poses, see predicted hydrogen bonds to the protein, and browse 2D structures aligned on the maximum common substructure. Search by name or SMARTS.
SAR browser for PXR. Click a parent hit to see its analogs sorted by pEC50, each aligned to the parent with the shared substructure highlighted. Parent cards summarize the potency distribution of their analogs.
108 PXR crystal structures in a Mol* viewer alongside a sortable table of 2D structures, pEC50, and Butina cluster. Check structures to overlay them, switch the protein between cartoon, backbone, and surface, and turn on pocket side chains or distance-based H-bonds. Filter by OCNT_ID, structure, or SMILES.
Generative topographic map of 7,298 molecules colored by LogS, the negative log of aqueous solubility in µM. Drag the cutoff sliders to change where the red/yellow/green split falls and watch the pies, histogram, and cards recolor. Tap a node to load its molecules into the structure grid.
Data from Fang, Cheng, et al. “Prospective validation of machine learning algorithms for absorption, distribution, metabolism, and excretion prediction: An industrial perspective.” J. Chem. Inf. Model. 63, 3263–3274 (2023). doi:10.1021/acs.jcim.3c00160
The eight blind PXR structures where the ligand does not sit the same way in the two copies in the asymmetric unit. With the protein superimposed, the chain A and chain B poses disagree by 3.2–10.2 Å. Each pair is drawn in a Mol* viewer with the A and B ligands colour-coded, and the protein and pocket side chains of either chain can be turned on independently – leave both proteins off to compare the ligands against the pocket residues alone. A reminder that “the” crystallographic pose is sometimes two different poses.