[SOUND]. In this lecture I would like to discuss

with you the strengths and limitations of the different types of models one can use

in systems biology. Let's start with statistical models.

Statistical models are very useful as bookends in, in defining biological

functions at a systems level. Especially sort of at very soft of distal

relationships, such as a gene and a disease, so but it can also be a gene and

phenotype. And the phenotype can be at a

sub-cellular, cellular or sort of tissue organ levels.

But let me give you the example from this gene and the disease because this is most

often used. And here is an example from what is a

genome, by association study. Where one can have a case, with a certain

type of disease. One can have a control one can get say

DNA from these patients with the disease and the control patients compare what

things change and make molecular correlates to the physiological

functions. And these kinds of correlations can

provide context to understanding how these genes may be involved or are likely

to be involved in the disease. Statistical models, like I said before in

a couple of the other lectures, statistical models, have the ability to

capture probabilistic relationships. Not all relationships or not all edges

between nodes whether there is gene to gene interaction protein-protein

interactions occur in a deterministic fashion which is that they are always,

that the two are always pressing back. There may be the interactions and hence

the edges are operational under a variety of conditions there so and hence have a

probabilistic relationship. And here is an example and statistical

models of the ability to capture these probabilistic examples relationships.

Here is an example that I took from a paper by Eric Schadt and his colleagues

where they devised a statistical test that would link a gene to a to a

phenotype. and this can be and this linkage can be

in the genus called a loci a locus or a position that a chromosome to a gene to a

sort of a phenotype or a clinical trait quality.

And they can be a variety of ways in which they can be linked using a variety

of statistical tests and they devised, in this paper devised a statistical test.

That sort of is built on a, a, assumption of causal relationships.

These kinds of statistical test, although they have, they're not very widely used

as yet tell us that even when you don't know all the details in between.

One can clearly link a gene to a to a phenotype and thus allow for building of

a, thus allow for an understanding of how the system might behave.

And have predictive capability of systems behaviors.

So those are the positives with regard to statistical models.

Now, like everything else, statistical models have their limitations, and one of

these limitations is that statistical models obscure mechanistic details.

Statistical models do not consider mechanisms, so for instance if you're

looking at the relationship between a gene and the level of messenger RNA in a

statistical correlation model. One wouldn't really consider whether the

change in levels is due to CNV copy number variation or DNA methalation, the

promoters and so on. So statistical models is just to

co-relate two nodes with respect to their levels or, with their respective level or

activities. but, not really, does, does not really

address mechanism. Statistical models, of course, do not

consider details of pathways of action in measuring relationships between genes and

phenotypes. So, what happens is that when you don't

consider these mechanistic details, sometimes you miss out on clues.

And these missing clues, often if they're modifiers of the phenotype can lead to

sort of limited. can lead to limitation in you ability to

predict whether a gene or changes in the certain gene or a lurkers will cause a

certain change in the disease. Because that may only be only valid under

certain conditions and not under other conditions.

So not considering the mechanistic details can lead the correlations to

fail. And then when it fails because you don't

have a knowledge and understanding of the mechanisms.

If one is not able to explain why the correlation doesn't work under a certain

circumstance. This is a limitation of statistical

models. The other, of course, limitation

statistical models is that, they do not provide dynamic or spatial information.

Statistical models test most assume a steady state relationship between two

entities. say for instance the gene.

Or a protein and the phenotype being produced, and so because of the but in

real life, things are changing all the time.

So for complex diseases, phenotypes change with respect to time and

treatment. such as, like if you treat somebody with

drug a cancer patient with the drug. They eventually develop resistance for

the drug so it is often necessary to capture the dynamic details of the

phenomenon in the model. And so the statistical models cannot

satisfy this requirement. For many biological functions also,

especially at the cellular level, spatial specification is essential for

understanding function. And spatial specification cannot be

represented easily in statistical models. So, this is also a limitation.

The next type of models one can think about are of course network models.

And network models are essential for understanding the topology of the system.

I I have taken here sort of a network graph of the subway system.

in lower Manhattan and Brooklyn, or actually Manhattan and Brooklyn, to to

tell you that if you want to get from say Bronx, which is up here, to Brooklyn

that's down here. You need to understand how the subway

systems, or you need to know how the subway systems run through Manhattan to

get from one borough to the other. Networks in some ways are similar to

subway maps because they can tell you how you can get from here to there.

That is from say genes to proteins and functions and phenotypes and what routes

can are, can be taken. And of course what the points of

connections are. So all of these are sort of important to

characteristics of the system. And so it is essential to understand,

topology or organization of the system, without which one doesn't really have an

understanding of the system as a whole. The other great advantage of network

models are they are flexible and can be used for multiscale descriptions of a

system. Since the user defines the edges this

allows for networks to be built across a single scale of interact, a single scale,

or between scales of interaction. So for instance, you can build a network

that is can where all the connectivities are at the level level of the

protein-protein or protein-DNA interactions.

You can also build a network where certain molecular entities are connected

to certain tissue or organ characteristics.

So this ability to build these networks of act within scales and between scales

allows for one to connect key reactions at one level to functions at another

level. This is actually very important,

especially in terms of drug actions. Because most drugs are small molecules.

They interact with their target at the protein level.

The effect of the drug is really seen in the tissue or organ or indeed the whole

organismal level. So having networks, both molecular

networks, the networks that propagate from cells to tissues to organs to whole

organisms is sort of very important aspect.

And network models are sort of useful for doing this.