Darwinian Evolution of Virtual Organisms

我们已经构建了一个虚拟有机体进化的模型。在该模型的最初版本中，“有机体”是布尔网络，其结构由可遗传但容易发生突变的“基因组”指定。每个网络都经历了一个“生命周期”，在这个“生命周期”中，“任务”是尽可能接近地生成由调查者指定的目标函数。选择生物体是因为它们在执行任务时的准确性。这导致了在模拟目标功能方面非常有效的生物体的进化。当生物体受到以注入的数字噪声形式的环境扰动时，它们在某些情况下可以通过发展排除噪声的能力来适应。然而，我们出人意料地发现，在其他情况下，生物体会通过将无关和无关的噪音特征融入到它们自己的结构中来适应，这样它就会被困在有机体的功能中，成为其功能所必需的。这一过程，我们称之为“噪音印记”，似乎类似于在文化进化过程中，外来因素被带入人类文化的许多情况。在当前项目期间，我们已将模型并行化并在Beowulf集群(NIH Bethesda的Biowulf和我们实验室的专用48处理器集群)上实施，将计算能力提高了2-3个数量级。利用这种额外的力量，我们正在构建一个“有机体”和“模因”(“文化”元素)共同进化的模型，后者通过在有机体中传播进行复制，并充当控制有机体与环境相互作用的“软件”，类似于人类基因组和人类文化的共同进化。我们假设，这种两个层次的进化过程可能会产生类似于有机体之间的“群体选择”的后果。我们推测，进化中的模因可能分为基于来自环境的经验信息而在短期内行动的“理性”模因和体现关于进化过程本身的永久的、不可观察的“真理”的“神话”模因，这些“真理”对支持有机体在多代时间尺度上的生存具有价值。这可能会提供一种与所有人类社会共存的“科学”和“宗教”文化元素的类比。我们正在特别关注在紧密耦合的有机体对中可能发生的各种共同进化相互作用，例如宿主/共生对以及基因和模因之间的类似关系。由于这样一对生物体有不同的生殖周期，尽管它们紧密地相互依赖，所以它们有“自由意志”以“重新谈判”它们的关系的方式进化。对于人类文化中发生的许多权力和主导地位的转移，这些转变似乎是一个有用的模型，我们正在寻找这些转变是否在计算模拟中表现出来。特别令人感兴趣的是，基因和文化的共同进化可能会推动生物进化朝着更长的生殖后寿命的方向发展(因为文化的教学不可能在一个生物生殖周期内完成)。我们将在共同进化模型中寻找类似的现象。 自上一份报告以来，我们开发了一个额外的模拟模型，专门研究非理性冒险行为的演变方式。在自然选择的随机模型中，在不增加(或甚至减少)其平均值的情况下“赌博”增加其适应度变化的有机体被发现处于劣势。然而，当文化或其他表观遗传的影响被包括在模型中时，这种模式发生了戏剧性的变化，因此对有机体没有直接好处的冒险行为仍然在多个世代中被强烈选择。这个模型的最初版本涉及无性有机体(或者相当于Y染色体上编码的特征)。风险模型现在正被扩展到完全的有性繁殖，我们正试图看看在抽象的布尔网络模型中是否存在同样的“文化”进化的一般效应。

We have constructed a model of the evolution of virtual organisms. In the initial version of the model, the "organisms" are boolean networks, whose structure is specified by a "genome" which is heritable but subject to mutation. Each network is stepped through a "life cycle" in which the "task" is to generate, as closely as possible, a target function specified by the investigator. Organisms are selected for their accuracy in carrying out the task. This resulted in the evolution of organisms that were highly effective at simulating the target function. When organisms were subjected to environmental perturbation, in the form of injected, numerical noise, they could, under some circumstances, adapt by developing the ability to exclude the noise. However, we unexpectedly discovered that, under other cirumstances, organisms would adapt by incorporating extrraneous and irrelevant features of the noise into their own structure, such that it became enmeshed in the function of the organism and necessary for its function. This process, which we dubbed "noise imprinting" appears to be analogous to many situations in which extraneous elements are entrained into human culture during the process of cultural evolution. In the current project period, we have parallelized the model and implemented it on Beowulf clusters (Biowulf at NIH Bethesda, and a dedicated 48-processor cluster in our lab) increasing computational power by 2-3 orders of magnitude. Using this added power, we are constructing a model of the co-evolution of "organisms" and "memes" ("cultural" elements) the latter of which replicate by spreading among the organisms and act as "software" to control the organisms' interactions with the environment, analogous to the co-evolution of the human genome and human culture. We hypothesize that this two level evolutionary process may have consequences analogous to "group selection" among the organisms. We speculate that the evolving memes may divide into "rational" ones that act in the short term on the basis of empirical infromation from the environment and "mythological" ones that embody permanent, non-observable "truths" about the evolutionary process itself that are of value in supporting survival of the organisms over multi-generational time scales. This might provide an analog of the "scientific" and "religious" cultural elements that co-exist in all human societies. We are giving particular attention to the kinds of co-evolutionary interactions that can occur in tightly coupled pairs of organisms, such as host/commensal pairs and the analogous relationship between genes and memes. Since organisms of such a pair have separate reproductive cycles, despite their close interdependence, so they have the "free will" to evolve in ways that "renegotiate" their relationship. We are looking to see if these transformations, which seem to be a useful model for many of the shifts in power and dominance that occur in human culture, are manifest in the computational simulations. Of particular interest is the possibility that the co-evolution of genes and culture may drive biological evolution in the direction of longer post-reproductive life span (because the teaching of the culture cannot be completed within one biological reproductive cycle). We will look for similar phenomena in the co-evolution models. Since the previous report we have developed an additional simulation model to study specifically the way in which irrational risk-taking evolves. In a stochastic model of natural selection, organisms that "gambled" increasing the variance of their fitness without increasing (or even decreasing)its avereage value were found to be at a disadvantage. However, when the effects of cultural or other epigenetic inheritance were included in the model, that pattern shifted dramatically, so that risk-taking behavior of no immediate benefit to the organism was nevertheless strongly selected over multiple generations. The original version of this model concerned asexual organisms (or equivalently, traits coded on the Y chromosome). The risk model is now being extended to full sexual reproduction, and we are trying to see if the same generic effect of "cultural" evolution exists in the abstract boolean network model.