Scoring¶
aizynthfinder is capable of scoring reaction routes, both in the form of MctsNode objects when a search tree is available,
and in the form of ReactionTree objects if post-processing is required.
Currently, there are a few scoring functions available
State score - a function of the number of precursors in stock and the length of the route
Number of reactions - the number of steps in the route
Number of pre-cursors - the number of pre-cursors in the route
Number of pre-cursors in stock - the number of the pre-cursors that are purchaseable
Average template occurrence - the average occurrence of the templates used in the route
Sum of prices - the plain sum of the price of all pre-cursors
Route cost score - the cost of the synthesizing the route (Badowski et al. Chem Sci. 2019, 10, 4640)
The State score is the score that is guiding the tree search in the update phase, and this is not configurable.
In the Jupyter notebook GUI one can choose to score the routes with any of the loaded the scorers.
The first four scoring functions are loaded automatically when an aizynthfinder object is created.
Add new scoring functions¶
Additional scoring functions can be implemented by inheriting from the class Scorer in the aizynthfinder.context.scoring.scorers module.
The scoring class needs to implement the _score_node, _score_reaction_tree and the __repr__ methods.
This is an example of that.
from aizynthfinder.context.scoring.scorers import Scorer
class DeltaNumberOfTransformsScorer(Scorer):
def __repr__(self):
return "delta number of transforms"
def _score_node(self, node):
return self._config.max_transforms - node.state.max_transforms
def _score_reaction_tree(self, tree):
return self._config.max_transforms - len(list(tree.reactions()))
This can then be added to the scorers attribute of an aizynthfinderfinder object. The scorers attribute is a collection
of Scorer objects.
For instance to use this in the Jupyter notebook GUI, one can do
from aizynthfinder.interfaces import AiZynthApp
app = AiZynthApp("config_local.yml", setup=False)
scorer = DeltaNumberOfTransformsScorer(app.finder.config)
app.finder.scorers.load(scorer)
app.setup()