Winner-take-all molecular network

Great article with a misleading title: "A synthetic protein-level neural network in mammalian cells" (Science Magazine: link free bioRxiv & medRxiv preprint: link)

The premise that cells are (fast) information-processing systems that classify molecular signals and act accordingly is a great start. The idea of using "de novo designed protein heterodimers and engineered viral proteases to implement a synthetic protein circuit that performs winner-take-all computation" is exciting.

Why bring "neural network" into the mix?

The winner-take-all molecular network does have the logic architecture of a simple perceptron and the authors show that their system has "tunable" decision boundaries. My issue with the title is that the "tunable" parameters (varying levels of certain plasmids) are not learnt but explored manually: imagine a neural network with training done by the user by manually testing "various" weight values. So what we have here is a logic circuit with no training. Hardly a neural network.