A Single-Layer Artificial Neural Network Type Architecture with Molecular Engineered Bacteria for Complex Conventional Computing

Artificial neural networks (ANNs) are inspired by the human brain and they are designed to compute complex problems bearing resemblance to the way in which the human brain does. Hardware implementation of artificial neural networks has been carried out in photonics, material-based neuromorphic chips and DNA strand displacement processes. However, it has never been physically realized in living cells. Here, we adapt the basic concept of ANN and develop the single-layer ANN-type architectures with engineered bacteria to perform complex computations. We have built and optimized molecular devices within the bacterial cells where they act like artificial neuro-synapses and the bacterium carrying the molecular device behave like an artificial neuron (Bactoneuron). The externally supplied input chemicals, associated with each bactoneuron get linearly combined along with their weights and the bias into the summation function which further gets processed through a non-linear activation function to give rise to the output in terms of the fluorescent protein production. We further illustrate a general framework to build bactoneuron-based single layer ANN-type architecture from the truth table of a given function without considering its electronic circuit design where each bactoneuron is employed to perform a specific function which is the subset of the final function and at the network level, their collective behavior gives rise to the entire final function in a multi-population culture. We have also developed the sets of rules to distribute a functional truth table into the smaller parts and design the bactoneurons out of them, adjusted their weights and biases manually by engineering the molecular interactions within the bacteria and finally observed their performances at the network-level. In this way, we have developed bacteria-based single-layer ANN type architectures working as a 2-to-4 decoder and a 4-to-2 priority encoder. This work may possess a new design approach for complex biocomputing.