Phenotypic models and automatic science for systems biophysics

在这个提议中，我扩展了我以前在生物进化模拟和免疫系统建模方面的工作，以开发旨在揭示生物学理论原理的新策略，类似于物理学中可以发现的原理。我将使用早期免疫识别作为感兴趣的主要系统，但开发的方法将足够通用，可以应用于其他地方。我将首先使用计算和统计物理学、信息和统计决策理论的工具，对免疫T细胞触发免疫反应的集体决策进行建模。然后，以我们提出的配体识别的“自适应排序”模型为基础，研究模型降阶的相关问题。用于模型简化的技术受到统计物理学的启发，特别是重整化，并且可以自动实现数值。我将联合收割机这些技术与我们以前设计的数字工具相结合，以开发一种“自动化科学”方法。我们的目标是从生物行为开始，首先生成预测生物系统的计算模型类，然后将它们自动简化为简化模型，我们可以从中提取简单的生物学原理。虽然这种方法是理论上的，但它是通过目前的实验合作建立在生物现实基础上的。* 我们的长期目标是产生一个“理论的理论”：我们过去已经建立了生成生物系统预测模型的方法，现在这个建议涉及模型生成本身的研究以及我们可以在模型之间建立的桥梁。有两个层面的影响：在生物学方面，这将跟进我们以前的工作，以提供更好的免疫策略的基本理解。在理论方面，我们的策略将有助于弄清楚如何超越目前在所谓的“系统生物学”中使用的“大数据/大模型”范式，从非常复杂的噪声生物数据集和实验中提取预测生物学原理。

In this proposal, I extend my previous work on simulations of biological evolution and modelling of immune system to develop new strategies aiming at uncovering theoretical principles for biology, similar to what can be found in physics. I will use early immune recognition as the primary system of interest, but the methods developed will be general enough to be applied elsewhere.***I will first model collective decision of immune T cells to trigger immune response, using tools from computational and statistical physics, information and statistical decision theory. Then, I will study the related problem of model reduction, using our model of "adaptive sorting" for ligand recognition as a basis. The techniques used for model reductions are inspired by statistical physics, in particular renormalization, and can be automatically implemented numerically. I will combine these techniques with the numerical tools we have designed previously to develop an "automatized science" approach. The goal is to start with biological behavior to first generate classes of computational models predictive of biological systems, then reduce them automatically to simplified models from which we can extract simple biological principles. While this approach is theoretical, it is grounded in biological reality through present experimental collaborations. ***Our long term ambition is to produce a "theory of theory": we have built methods to generate predictive models of biological systems in the past, and this proposal is now concerned with the study of models generation itself and on the bridges we can build between models. There are two levels of impact: on the biology side, this will follow up on our previous work to provide much better fundamental understanding of immune strategies. On the theoretical side, our strategy will help figuring out how to go beyond the "big data/big model" paradigm currently used in so-called "systems biology" to extract predictive biological principles from very complex noisy biological datasets and experiments.**