Forward prediction issue with large reagents (ligands)
I'm not sure this is necessarily a software bug, but I think can be chalked up to undesired behavior. See screenshot below for example.
For this contrived example Suzuki reaction, ASKCOS does a decent job at recommending a Pd catalyst/ligand, however when evaluating with the forward predictor, I suspect the largest molecule is identified as the major product, even if the correct carbon-carbon bond formation was indeed correctly predicted.
I believe this unwanted behavior has existed for quite a while (https://gitlab.com/mlpds_mit/ASKCOS/askcos-core/-/blob/dev/askcos/synthetic/evaluation/template_free.py#L100), but it might be worth addressing at some point.