BIC: Evolving Signal Processing Circuits from Biological Reaction Networks

这个项目的目的是了解生物有机体如何处理信号，以及苏坎的理解如何影响人造设备的未来发展。我们通常将计算与人造设备联系在一起，特别是无处不在的数字计算机或鲜为人知的模拟计算机。然而，有计算设备离家更近。所有的生物有机体几乎都持续地暴露在各种各样的环境变化和冲击中。为了在这种变化中生存下来，生物体已经进化出位于生物体外表面的传感器蛋白，它可以检测到环境变化的所有方式。这些传感器蛋白依次连接到由细胞内相互作用的蛋白质组成的所谓的信号网络。这些信号网络负责根据所有的感官输入做出适当的决定。在这个阶段，我们还不太了解这些网络执行的决策处理类型。我们使用从数字计算机到模拟设备的各种技术来控制我们的机器。在过去的50年左右的时间里，复杂的人工控制装置的设计已经成熟到几乎所有的装置都内置了某种控制系统的程度。我们之所以如此擅长设计人工控制系统，是因为我们对控制的基本理论原理有着透彻的掌握。我们用来制造控制装置的主要技术是基于电子学的。走进任何一家书店，你都会发现电子设计方面的书籍，但在生物控制系统方面，我们没有一本电子设计手册。因此，我们对生物控制系统如何工作，如何执行决策以及生物控制的基本原则知之甚少。我们的方法是在数字计算机上进化，人造生物信号网络。根据网络设计执行的任务，我们进化出最接近实现这一目标的网络。例子包括进化一个网络，它可以对环境中的突然变化保持鲁棒性，或者反过来进化一个网络，它可以快速响应环境变化。此外，其他目标将包括电子学中常用的信号处理技术，例如，我们可能会进化出一个可以振荡的网络或一个可以进行某种运算的网络。通过这些方法，我们将产生生物类网络，它将有能力执行所有常见的电子信号处理任务。最终的结果将是一个大型的网络库。从这个库中，我们将对网络进行逆向工程，以了解它们如何完成进化的任务。最后，我们将这些网络与真实的生物网络进行比较，看看我们是否能找到等效的"设计"。最终目标是编写生物信号控制网络的"电子"设计手册。我们在这个应用中提出的工作影响了许多科学领域。它结合了分子生物学、计算机科学、控制理论、进化算法、信号处理和电路理论的工作，工程科学将从这项工作中受益，因为它能够检查在分子水平上进行的信号处理的例子，而生物科学将从理解真实的生物网络的基本控制原理中受益。此外，基于分子的电路必须处理噪声（这在工程科学中被广泛处理），这项工作可能对基于纳米技术的控制系统的实现有重要影响。

The aim of this project is to understand how biological organisms process signals and how suchan understanding might impinge on the future development of man-made devices.We normally associate computation with man-made devices, particularly the ubiquitous digitalcomputer or the lesser-known analog computer. However there are computational devices muchcloser to home. All biological organisms are exposed almost continuously to a huge variety ofenvironmental changes and shocks. In order to survive such changes, living organisms haveevolved sensor proteins located on the outer surface of the organism which can detect all mannerof environmental changes. These sensor proteins are in turn connected to so-called signalingnetworks composed of interacting proteins inside the cell. These signaling networks areresponsible to making an appropriate decision based on all the sensory inputs. What is not wellunderstood at this stage is the type of decision processing that is carried out by these networks. Inman made devices we employ a variety of techniques from digital computers to analog devicesto control our machines. Over the last fifty years or so, the design of sophisticated man-madecontrol devices has matured to the extent that almost all devices now have some kind of controlsystems built into them. The reason why we are so good at designing artificial control systems isthat we have a thorough grasp of the underlying theoretical principles of control. The primarytechnology that we used to build control devices is based on electronics. Walk into any bookstore and one will find books on electronic design.In relation to biological control systems we do not have the equivalent of an electronics designhandbook. As a result we understand very little about how biological control systems work, howthey carry out decisions and what the underlying principles of biological control are. Ourapproach is to evolve on digital computers, artificial biological signaling networks. Dependingon what task the network is designed to perform, we evolve networks which will come closest toachieving this objective. Examples include evolving a network which can be robust to suddenchanges in the environment, or conversely evolving networks which can quickly respond toenvironmental changes. In addition other objectives will include common signal processingtechniques employed in electronics, for example we might evolve a network that can oscillate ora network that can carry out some arithmetic. By these means we will generate biological likenetworks which will have the capacity to carry out all the common electronic signal processingtasks. The end results will be a large library of networks. From this library we will then reverseengineer the networks to understand how they accomplish their evolved tasks. Finally we willcompare these networks to real biological networks to see if we can find equivalent 'designs'.The ultimate goal is to write the 'electronics' design manual of biological signaling controlnetworks.The work we propose in this application impinges on many areas of science. It combines workfrom molecular biology, computer science, control theory, evolutionary algorithms, signalprocessing and electrical circuit theory.The engineering sciences will benefit from this work by being able to examine examples ofsignal processing carried out at the molecular level and the biological sciences will benefit by anunderstanding of the underlying control principles of real biological networks. In addition,molecular based circuitry has to deal with noise (which is dealt with extensively in theengineering sciences), this work may have an important bearing on the implementation ofnanotechnology based control systems.