Pan-substructural hierarchical algorithm for extensive chemical space analysis

A computer-implemented method for constructing a hierarchical library of molecular substructures for synthetic accessibility evaluation, the method comprising:

• defining a zero layer comprising a core structure represented as a SMARTS string;
• defining an element alphabet comprising atomic primitives for heteroatoms;
• generating a first layer by combinatorially replacing generic atom symbols in the zero layer SMARTS string with atomic primitives from the element alphabet to produce heterocycle type substructures;
• defining a substituent alphabet comprising bond primitives and atomic primitives for substituent types;
• generating a second layer by replacing attachment point symbols in the first layer substructures with hydrogen atoms and substituent placeholders to produce position-specific substructures;
• generating a third layer by enumerating substructures having two or three substituent positions;
• generating a fourth layer by combinatorially substituting all substituent types from the substituent alphabet at each substituent position in the third layer substructures;
• collecting occurrence statistics for each substructure in the fourth layer by querying a reference dataset of synthesized compounds;
• identifying synthetically irrelevant substructures as those having occurrence frequency below a predetermined threshold;
• and
• storing the hierarchical library comprising the zero layer, first layer, second layer, third layer, and fourth layer with associated occurrence statistics.

A computer-implemented method for filtering molecular structures generated by a generative chemistry model, the method comprising:

• receiving a batch of molecular structures from a generative chemistry model, each molecular structure represented as a SMILES string;
• for each molecular structure in the batch:
• parsing the SMILES string to generate a molecular graph object;
• applying a hierarchical filtering algorithm to determine whether the molecular graph object contains a synthetically irrelevant substructure by traversing a hierarchical library of SMARTS-encoded substructures;
• assigning a PASS status when no synthetically irrelevant substructure is detected;
• and
• assigning a FILTER status when a synthetically irrelevant substructure is detected;
• generating a filtered set comprising molecular structures having PASS status;
• and
• outputting the filtered set for synthesis prioritization.

A computer-implemented method for selecting a synthetic route for a target molecule, the method comprising:

• receiving a target molecule represented as a SMILES string;
• generating a plurality of candidate synthetic routes for the target molecule;
• for each candidate synthetic route:
• identifying intermediate molecular structures;
• applying a hierarchical filtering algorithm to each intermediate molecular structure to determine presence of synthetically irrelevant substructures;
• calculating a synthetic accessibility score based on number of intermediate molecular structures containing synthetically irrelevant substructures;
• selecting a preferred synthetic route having a lowest synthetic accessibility score;
• and
• outputting the preferred synthetic route for synthesis execution.