Finding the ‘Effective Variables’ in Biological Systems

Top

00:00:08.00 Hi, my name is Rama Ranganathan, I'm a Professor at the Green Center for Systems Biology
00:00:12.00 and affiliated with the departments of physics and pharmacology
00:00:15.00 at UT Southwestern Medical Center
00:00:17.00 And I'll be doing a talk in this series
00:00:21.00 called "great questions in life sciences, intersection between biology, computation, and physics
00:00:25.00 Specifically, the problem I would like to ask
00:00:28.00 is a very interesting one, it's about finding the effective variables
00:00:33.00 in biological systems. So what I like to do is to describe the nature of this problem, to tell you why
00:00:39.00 finding these effective variables is an important yet a difficult problem
00:00:44.00 I'll try to propose an approach that leads to specific problems of interest
00:00:49.00 people coming from quantitative sciences into biology. So a grand goal in biology
00:00:54.00 is to quantitatively explain the behavior and evolutionary origin of so called
00:00:59.00 complex systems at many scale. I'm showing biological systems at a variety of scales
00:01:05.00 from the atomic scale, that is the structure and function of proteins
00:01:10.00 to things at the cellular scale, the designate of cellular macromolecules and systems
00:01:15.00 like the flagellum, to the design of cells and tissues
00:01:21.00 and ultimately the design and principles of ecosystems
00:01:24.00 the goal is to explain the behavior and the origin at all these scales
00:01:29.00 but the problem is difficult for 3 reasons. The first is
00:01:32.00 the problem of heterogeneity. In all these scales, system components, the things that make the parts of
00:01:39.00 make up these systems are highly heterogeneous where some of the parts and connections
00:01:43.00 are much more important than others.
00:01:45.00 For example, we know that in proteins, not all of the amino acids act equally
00:01:49.00 In fact, there a heterogeneity in the importance of amino acids
00:01:52.00 towards the central properties of proteins such as foldability, biochemical functions, and so on
00:01:58.00 the same thing is true at a subcellular systems and a macromolecular systems
00:02:04.00 the importance of the parts is heterogeneous and often is not obvious
00:02:09.00 when looking at those systems which are the parts that are important, which are the parts that are less important
00:02:13.00 the second problem is that of cooperativity
00:02:15.00 that is the systems are made up of parts, but the parts work in a way that cannot predict the whole
00:02:21.00 cannot be predicted in the activities of the parts when calculated individually
00:02:25.00 there is also potentially nonlinearity in the system
00:02:31.00 so that the behavior of the system taken overall is not an obviously decomposition into the activity of the parts taken individually
00:02:39.00 the third problem is the problem of modularity in that different groups of a component can be grouped in a quasi-independent way
00:02:47.00 to provide different aspect of function at all these scales
00:02:52.00 So the existence of the properties, the facts that these parts that make up these systems
00:02:57.00 the fact that these amino acids make up proteins, proteins and small molecules that make up macromolecular complexes and cells
00:03:03.00 cells that make up tissues, and species that make up ecosystems
00:03:08.00 the fact that these component descriptions involves heterogeneous, cooperative groups imply that what we need is a
00:03:18.00 fundamental re-parameterization of these systems to redo the
00:03:21.00 fine units of function or what I call the effective variables
00:03:25.00 as not these parts, but groups of these parts that act cooperatively, and ideally ranked by their relevance
00:03:32.00 the question is how can we do that?
00:03:35.00 The goal in doing this is to provide some sense of what natural description of what a system is. It may be intuitive
00:03:45.00 think about a protein as made up of amino acids, but in terms of the function of the protein,
00:03:51.00 It may not be the actual description. Groups of amino acids work cooperatively may be a more natural description
00:03:57.00 of what the system is. And in some sense, one could argue that solving this problem, providing a more natural description of what a system is
00:04:05.00 should be a necessary preclude for mechanistic studies of how it works
00:04:09.00 another words, understanding how it works may be simpler if we understand the effective variables that make up a system
00:04:17.00 The question is how can we do this practically?
00:04:20.00 Well it turns out with the massive expansion of techniques for massive data collection
00:04:25.00 for all these scales, people are collecting methods for collecting many examples of protein structures
00:04:32.00 and protein sequences and measurements of protein function
00:04:36.00 of course, many are working at the cellular scale to characterize many different cell types under the background of perturbation for example mutations
00:04:43.00 and small molecule perturbations
00:04:47.00 and even at the ecosystem scale, more recently people have developed synthetic ecosystem laboratories
00:04:51.00 with many replicates so people can look at the system's behavior at that scale
00:04:59.00 so with the ability to collect data in nature. The proposal is to look not at individual systems
00:05:08.00 but at statistics and many realizations of this system to see what is conserved or invariant over the ensemble
00:05:16.00 fundamentally, the variants of properties of an ensemble of realizations of a system
00:05:23.00 is itself a relevance in systems components and reactions
00:05:27.00 things that are invariant over an ensemble tends to be relevant
00:05:31.00 the second thing to do since we are interested in the cooperativity of components
00:05:34.00 is to look at correlations and epistasis between the conserved system components and reactions
00:05:42.00 and this is a basis, such a correlational analysis is a basis for decomposing into the proper cooperative modular units
00:05:50.00 that is the effective variables. Let me give an example of this, at the scale of atoms that we've been talking about.
00:05:58.00 but the approach should be more general than just this case study
00:06:03.00 So here is an example where the protein can be thought of as made up of a large collection of amino acids
00:06:09.00 that interact together, but to understand which amino acids are relevant and what are the cooperative units
00:06:15.00 this kind of approach, one collects an appropriate statistical ensemble
00:06:19.00 of instances of a protein, in this case, a protein family that shares the property of folding into that structure and carrying out the biochemical activities
00:06:29.00 and from such a statistical ensemble, one can make a matrix of relevant correlations of all the amino acids that make up the protein
00:06:39.00 and with that correlational matrix at hand
00:06:43.00 one can apply techniques such as structural decomposition
00:06:47.00 very simple way of information content in correlation matrices
00:06:51.00 to deduce collective groups of amino acids that now more appropriately represent the determinants
00:06:58.00 of protein structure and function. So this is an example at the atomic scale
00:07:02.00 but I want to point out that there are non-trivial quantitative problems at every stage of analysis
00:07:09.00 in devising this point of approach more generally for all biological complex systems
00:07:14.00 and let me just illustrate the nature of the problems that need to be worked on
00:07:18.00 For example it's not obvious when you take a statistical approach how to make an ensemble that represents
00:07:24.00 the properties of interest. In case of proteins, it seems nature that a protein family
00:07:30.00 that shares the function of folding and function and an evolutionary history
00:07:36.00 should be an appropriate ensemble over which the statistics is computed
00:07:40.00 but it's not necessarily obvious in many other systems, at a cellular scale, at the scale of ecosystems and so on
00:07:47.00 what should be an appropriate ensemble of data to collect
00:07:52.00 from which statistics can reveal the underlying variables of a system
00:07:57.00 that's a working problem to solve. The second problem is what type of statistical system one should apply
00:08:03.00 to make measures of relevance and correlation given an ensemble and the reason for this is
00:08:11.00 it depends on how the ensemble is collected for protein sequences. one of the complications for making appropriate statistics is that
00:08:19.00 the sequences are not randomly selected from the space of all sequences
00:08:24.00 these sequeces have a relationship to each other that comes from the phylogeny of all the species that contain that sequences
00:08:31.00 in other words, sampling is not random, but sampling is on some phylogenetic tree, and in such instances
00:08:39.00 it not necessary obvious what are appropriate statistical methods that define the relevance of the outcomes at positions or correlations between positions
00:08:50.00 So in general, one needs to work on what are the statistical methods that are appropriate to represent a statistical ensemble
00:08:57.00 the final problem is, how to decompose these correlational matrices to deduce the effective variables
00:09:05.00 in the simplest case, one apply linear methods such as spectral decomposition
00:09:10.00 but it's not so obvious that those decompositions are good decompositions of this information. In general, information in these correlational
00:09:20.00 in biological systems is hierarchical in nature. And for a hierarchically embedded system,
00:09:26.00 it's not yet clear what are good methods are in finding the effective variables
00:09:32.00 so these are problems are ripe to work on now for people coming into biology from quantitative sciences
00:09:39.00 because if these sort of questions can be addressed in general, then we have the possibility leading to really new understandings
00:09:50.00 of the so called design principles of complex biological systems
00:09:54.00 to understand the effective variables underlying these systems, as I said a preload
00:09:59.00 to ultimately deep mechanistic studies of these systems
00:10:04.00 so this is an example of a problem that is appropriate and ripe today for people in physics and quantitative sciences
00:10:13.00 coming into biology. Thank you very much!

This Talk

Speaker: Rama Ranganathan

Audience:

Educators of Adv. Undergrad / Grad
Researcher
Educators

Recorded: May 2014

Talk Overview

Biological systems span an enormous range of scales – from proteins to cells to tissues to ecosystems. Rama Ranganathan asserts that biological systems are difficult to understand because their components are heterogeneous, cooperative, and modular. He identifies three major questions facing computational biologists as they try to mathematically model the behavior and evolution of biological systems.

Speaker Bio

Rama Ranganathan

Rama Ranganathan received his B.S. in Bioengineering from the University of California, Berkeley and his Ph.D. and M.D. degrees from UC San Diego where he studied signal transduction in the invertebrate visual system. He was a post-doctoral fellow at Harvard Medical School, where he studied K+ channels, and at The Salk Institute, where he learned… Continue Reading