Abstract
Embedding computation in molecular contexts incompatible with traditionalelectronics is expected to have wide ranging impact in synthetic biology,medicine, nanofabrication and other fields. A key remaining challenge lies indeveloping programming paradigms for molecular computation that arewell-aligned with the underlying chemical hardware and do not attempt toshoehorn ill-fitting electronics paradigms. We discover a surprisingly tightconnection between a popular class of neural networks (Binary-weight ReLU akaBinaryConnect) and a class of coupled chemical reactions that are absolutelyrobust to reaction rates. The robustness of rate-independent chemicalcomputation makes it a promising target for bioengineering implementation. Weshow how a BinaryConnect neural network trained in silico using well-foundeddeep learning optimization techniques, can be compiled to an equivalentchemical reaction network, providing a novel molecular programming paradigm. Weillustrate such translation on the paradigmatic IRIS and MNIST datasets. Towardintended applications of chemical computation, we further use our method togenerate a CRN that can discriminate between different virus types based ongene expression levels. Our work sets the stage for rich knowledge transferbetween neural network and molecular programming communities.