Supervised learning in DNA neural networks.

Learning enables biological organisms to begin life simple yet develop immensely diverse and complex behaviours. Understanding learning principles in engineered molecular systems could enable us to endow non-living physical systems with similar capabilities. Inspired by how the brain processes information, the principles of neural computation have been developed over the past 80 years, forming the foundation of modern machine learning. More than four decades ago, connections between neural computation and physical systems were established. More recently, synthetic molecular systems, including nucleic acid and protein circuits, have been investigated for their abilities to implement neural computation. However, in these systems, learning of molecular parameters such as concentrations and reaction rates was performed in silico to generate desired input-output functions. Here we show that DNA molecules can be programmed to autonomously carry out supervised learning in vitro, with the system learning to perform pattern classification from molecular examples of inputs and desired responses. We demonstrate a DNA neural network trained to classify three different sets of 100-bit patterns, integrating training data directly into memories of molecular concentrations and using these memories to process subsequent test data. Our work suggests that molecular circuits can learn tasks more complex than simple adaptive behaviours. This opens the door to molecular machines capable of embedded learning and decision-making in a wide range of physical systems, from biomedicine to soft materials.