Distributed information encoding and decoding using self-organized spatial patterns

Biology can generate distinct self-organized patterns according to different initial conditions, and one could infer the corresponding condition given a pattern. Moreover, under the same or similar conditions, these patterns share global similarity but vary in detail due to random noise. Here, we leverage the above properties of bacterial colony patterns and combine with machine learning (ML) to achieve distributed information encoding and decoding with guaranteed security. Specifically, to encode, a message is converted into cell seeding configuration followed by colony growth, during which a colony pattern develops; to decode, we input the pattern into a trained CNN to convert it back to the original message. By modulating the patterning dynamic and encoding setup, we could tune the trade-off among encoding capacity, decoding accuracy and security, characterized by ML decoding performance. We also implemented ensemble techniques for enhancing decoding reliability and making full use of the expensive-to-obtain patterning data, and combined the framework with established cryptography techniques (e.g., encryption and hashing) to further enhance the security. Our method is applicable for a wide variety of pattern-formation systems and demonstrates a novel way of utilizing biological noise, as well as quantifying the extent of convergence for dynamical systems by using ML.