Zhiping (Peter) Zhang
Multistep Metabolic Pathway Design Using Deep Learning
In silico tools are indispensable for efficiently exploring and generating novel hypotheses in metabolic pathway design. However, while many computational frameworks have been proposed, very few successful examples of algorithm guided retrobiosynthesis have been reported. Inspired by recent progress of machine learning in computational chemistry, we explored the idea of combining deep learning-based ranking/pruning models with enzymatic template-based network expansion, in the hope of producing better in silico multistep metabolic pathway designs.