Analog and Stochastic Computation in Living Cells and Supercomputing Chips
Despite more than 15 years of research, synthetic circuits in living cells have been largely limited to a handful of digital logic gates and have not scaled. We show that one important reason for this failure to scale is an overemphasis on digital abstractions rather than on recognizing the true noisy, analog, and probabilistic nature of biological circuits. We show that synthetic and natural DNA, RNA, and protein circuits in cells must use analog, collective analog, probabilistic, and hybrid analog-digital computational approaches to function; otherwise, even relatively simple computations in cells will exceed energy, molecular-count, and cellular-resource budgets.
Analog circuits in electronics and molecular circuits in cell biology are also deeply connected: There are astounding similarities between the equations that describe noisy electronic flow in sub-threshold transistors and the equations that describe noisy molecular flow in chemical reactions, both of which obey the laws of exponential thermodynamics. Based on these similarities, it is possible to take a principled approach to design circuits in living cells. For example, we have engineered logarithmic analog computation in living cells with less than three transcription factors, almost two orders of magnitude more efficient than prior digital approaches to create a 'bio-molecular slide rule'. In addition, highly computationally intensive noisy DNA-protein and protein-protein networks can be rapidly simulated in mixed-signal supercomputing chips that naturally capture their noisiness, dynamics, and non-modular interactions at lightning-fast speeds via log-domain 'cytomorphic' circuits. Such an approach may enable large-scale design, analysis, simulation, and measurement of cells. To realize the promise of synthetic biology and systems biology for medicine, biotechnology, agriculture, and energy, we will need to go back to the future of computation and design and implement circuits via a collective analog approach like Nature does.
The implications of our results are general and apply to electronic, molecular, and neurobiological circuits: Neural and molecular circuits share 13 similarities from a hybrid analog-digital point of view. Thus, neural and neuromorphic circuits may be viewed as special cases of cellular and cytomorphic circuits that operate at more rapid time scales.