Modular RADAR

The natural immune system (NIS) searches for pathogens and responds by producing antibodies to clear them. The search for pathogens is through the physical space

of the body. The NIS has to search through a trillions of cells and find small amounts of pathogens. This search for a "needle in a haystack" gets even more difficult in

larger animals. For example, the search process is much harder in elephants (weight 500 kilograms) than in mice (weight 30 grams). The response is also a harder problem

in larger animals. Antibodies get diluted in blood, and larger animals have more blood. This implies that larger animals need to secrete more absolute quantities of antibody

than smaller animals to reach the same concentration of antibody in blood.

We recently showed (paper1, paper2, paper3) that despite being faced with a harder search and response problem in larger animals, the NIS is still able to search and respond in

almost the same time in larger animals as in smaller animals. For example, animals ranging from house finches (weight 3 grams) to horses (weight 300 kilograms) when

experimentally infected with West Nile Virus (WNV) can search for the pathogen and respond by producing antibodies within 2 - 4 days post infection. We call this scale-invariant

detection and response.

So how is this possible? How can you search ever larger spaces in the same time? Sounds hard right? The key is in the architecture of the NIS. The NIS has a decentralized detection network composed of lymph nodes. Lymph nodes are a region of tissue in which NIS cells and pathogens can interact with each other in a small volume. Here is a diagram of a lymph node and what is going on within and outside it.

Fig. 1. The schematic of a lymph node and the surrounding tissue which it services (called the

draining region). A kind of NIS cell called dendritic cell searches for pathogens in the draining

region and upon finding it, ingests it and migrates to to the nearest lymph node. Once in the

lymph node, it tries to "present" the pathogen to another NIS cell called a T helper cell. Only

one in a million of these T helper cells are specific to a particular pathogen, and hence only a

few of them can recognize it. Hence there is a very difficult search process both inside and

outside the lymph node. Once T helper cells recognize the pathogen, they start replicating and

building up an army. The lymph node also starts recruiting NIS cells from blood through high

endothelial venules (HEVs) which are vessels which bring in cells from blood.

This enormously speeds up the search process. We hypothesize that the NIS is capable of scale-invariant detection and response because it has a sub-modular RADAR

(Robust Adaptive Decentralized Automated Response) architecture of lymph nodes. What this means is that the lymphatic architecture is not only decentralized but sub-modular, i.e.

in larger animals lymph nodes are more numerous and bigger (as shown in the figure below).

Fig. 2. Left Panel: A hypothetical small animal with four lymph nodes. Right Panel: A larger hypothetical animal four times

larger than the smaller animal, with nine lymph nodes but each of them bigger than the ones in the smaller animal.

The red arrow shows the average distance that an antigen loaded dendritic cell has to migrate to reach the nearest

draining lymph node. The size of the incoming blue arrow into a lymph node depicts the size of high endothelial

venules (HEV) which feed into it and bring NIS cells into it.

The basic idea is very simple: a lymph node has two costs associated with itself:-

a) Local communication cost which is the time taken by antigen loaded dendritic cells to migrate to the nearest draining lymph node. If an animal has lots of small lymph nodes, then the region of tissue each of them is in "charge of", the draining region will be smaller and antigen loaded dendritic cells have to travel a smaller distance to get to the lymph node.

b) Global communication cost which is the time taken by the infected site lymph node to recruit NIS cells from blood through its high endothelial venules (HEV; they feed the lymph

node with NIS cells from blood). If a lymph node is large, then it can have more HEVs and can recruit NIS cells faster.

Hence there is a tradeoff between local and global communication. Having lots of small lymph nodes means each of them will have a small local communication cost but higher

global communication cost (since small lymph nodes will have fewer HEVs). On the other hand, having a few large lymph nodes means that each of them will have low global

communication cost but higher local communication cost (since there are now fewer of them and each is in charge of a larger draining region and hence antigen loaded dendritic

cells have to travel longer to get to the draining lymph node. You can hear me talk about this here and the slides are here.

To belabour the point, the optimal architecture is not one where the number of lymph nodes grows linearly with animal size and the size of lymph nodes remains the same in all animals (completely modular architecture) or where the number of lymph nodes remains the same in all animals but there size increases linearly with animal size (non-modular architecture). Rather the optimal architecture is one in which both lymph node numbers and their size increases with animal size (sub-modular architecture).

The following papers (paper1, paper2) discuss them in detail.

Applications to Distributed Systems

This is a great idea which the NIS has evolved. But we can also adapt it for our purposes! Distributed systems have becoming ubiquitous- from peer-to-peer systems to mobile ad-hoc networks, multi-robot control systems, intrusion detection systems and sensor networks, to name a few. Most distributed systems also have tradeoffs between local and global communication, just like the NIS. For example, see the multi-robot control system depicted below.

Fig. 3. Left Panel: a scaled down version of the multi-robot AIS system. The shaded regions are artificial lymph nodes (computer servers) and the unshaded regions are the artificial draining region. Light arrows denote communication between robots and servers (local communication) and bold arrows denote communication between servers (global communication). (B) Right Panel: a scaled up multi-robot AIS system with sub-modular architecture. Note that the number of artificial lymph nodes and their size (the number of robots they service and the number of software agents they have in memory) both increase with the size of the system.

The idea is that a computer server (an artificial "lymph node") is in charge of an artificial "draining region". The draining region has mobile robots. Whenever the robots encounter an obstacle, they communicate with their designated lymph node to get instructions on how to circumvent the problem. So their is local communication between a computer server and the mobile robots in its draining region. Each computer server also communicates with other servers to exchange high quality solutions. Thus there is also global communication.

Hence if there both local and global communication have bottlenecks (or example, there are bandwidth limitations) then the optimal architecture is one in which the number of computer servers and their size (the number of mobile robots each server must be in charge of and hence the amount of memory in a server) should both increase as the system grows bigger (see Fig. 3B).

Applications to Intrusion Detection Systems

We also applied our work to intrusion detection systems like LISYS and process Homeostatis (pH). You can read about our work here. I have some slides

and a talk on this subject.

Instructions on running the Agent Based Model

The ABM model is here: definition file and init file.

I use a very versatile ABM called CyCells available for free download here.

To run this in CyCells, install CyCells in a folder and paste these 2 files in the same folder.

Then run the following command:

./CyCells -d mys29.def -i mys29.init -t 10000

The -t specifies the number of time steps that it has to run.

The simulation produces a file (test.history) which has the data from the run.

Screenshot of the Agent Based Model

Here is a screenshot of the Agent Based Model. Green - draining lymph node, Red - immune system cells, Grey - virus