Map a molecular graph (the same { atoms:[{element}], bonds:[{a,b,order}] } format as mol_graph_validity) to a FIXED-LENGTH descriptor vector and measure a proper distance between two molecules.
Map a molecular graph (the same { atoms:[{element}], bonds:[{a,b,order}] } format as mol_graph_validity) to a FIXED-LENGTH descriptor vector and measure a proper distance between two molecules. The descriptor concatenates exact integer blocks - element counts, bond-order counts, and a Weisfeiler-Leman (Morgan-style) canonical subtree-color histogram - with a float spectral block from the normalized graph Laplacian eigenvalues. distance() is a weighted Euclidean metric (d>=0, d(x,x)=0, symmetric, triangle inequality) and interpolate() is a nearest-neighbor descriptor-path STUB. HONEST + LOAD-BEARING: this is a defined geometric embedding; a genuine GMDBS mesh57/HVP embedding was built and benchmarked and is information-equivalent to standard methods (no chemical advantage). Geometric distance here is NOT chemical similarity - now tested: it does not track composition; 3D shape resolves stereochemistry that graphs cannot, but the substrate only matches standard methods; no benchmark against a learned model has been run. GTI is EXACT in integer-only mode and an honest measured PRECISE when the float eigensolve is included; the verdict describes the COMPUTATION, not chemical truth.
Open it inside GDBS to save runs to Sandbox, attach results to a Worklog case, or share through a Gate client portal - all in the browser, nothing uploaded to anyone’s cloud.