Abstract: A method for training a model to estimate synthetic accessibility may be provided. A retrosynthesis-related synthetic accessibility model may be provided. A molecular structures database may be accessed. At least one molecular structure may be obtained from the molecular structures database with the model. The at least one molecular structure may be virtually sliced into synthon-like fragments with the model. A frequency of the synthon-like fragments in natural molecules may be determined with the model. Molecular descriptors for the synthon-like fragments may be calculated with the model. An aggregated synthetic accessibility score for the synthon-like fragments may be determined with the model. The aggregated synthetic accessibility score for the synthon-like fragments may be stored in a database for the model.