This September, I went to my first international conference in years, mainly because it dropped on my hometown of Melbourne. The Interntaional Conference on Systems Biology turned out to be extraordinarily good. If nothing else, it was worth it for Leroy Hood's majestic one-and-a-half hour keynote.
Here is a brief survey of my impressions.
Definition of Systems Biology
Being new to systems biology, er I mean proteomics, it was with some consternation that much of the conference was concerned with defining what the damn term meant. In hindsight, I realize this is a good thing - it means that the organizing principles are still up for grabs. I encountered three ways of defining systems biology:
- Roland Elis, a bioinformatician from EMBL, defined systems biology as what happens when you auto-generate a lot of data.
- Leroy Hood described systems biology as when one studies a network of interactions that give rise to a biological property.
- A bunch of modelers talked about systems biology as systems of differential equations.
Whole Cell Biology
I haven't really paid much attention to stem cells until this conference, but, fuck, they are a huge deal. The work on stem cells owe everything to Nobel prize winner Shinya Yamanaka - who discovered the genes that force mature cells to regress back to stem cells. What an epochal discovery! This gift has given a monstrous tool to the field of development biology, as researchers such as Andras Nagy, explore the myriad of factors involved in cell differentiation.
It's been 40 years since Richard Nixon announced the "War on Cancer", but we're no closer today to beating this disease. However, we may finally be getting a clear view of the terrain that we've been trudging on. For instance, we have now articulated the (hopefully) complete MAPK/ERK pathway, a key pathway involved in cancer that connect a whole bunch of receptors to the nucleus.
Cancer is now understood as the failure of the network, and not any one component. Now that the network has been mapped, it is the time for modellers to model the dynamics, ostensibly to see how the network reacts to drugs and fluctuations in reactants. I saw several talks from modelers such as Hans Westerhoff and Roland Elis, attempt to model the kinetic response of MAPK/ERK pathway. This involves the solution of hundreds of differential equations, each of which need to be carefully parameterized from experiment, or guesstimated by postdocs.
Talking about pathways, for model organisms such as yeast, groups such as that led by Brenda Andrews have attempted to map the entire network of pathways. They built two-hybrid knock-down for every known gene in yeast. Of course, much depends on the assay to measure the interaction, in this case, it is a cell-growth assay but the result is a provisional map of the entire interaction pathway in yeast.
Systems Biology Modeling
As a professional modeller of protein structures, I got quite excited by the plethora of talks on modeling. However, in those talks, I found that I was more confused than anything because the speakers often didn't, ur, explain what exactly they were modelling. How professionally inconsiderate.
I did finally figure it out. It seems in systems biology, models are mostly concerned with coupled chemical kinetic equations, usually in the form of ordinary differential equations, but also include master equations, and even boolean networks, a throwback to Stuart Kauffman's work from the 80's. These include some truly monstrous models involving hundreds of equations.
The system biology modeling community has built out some impressive infrastructure. For instance, Michael Hucka talked about a standard format that modelers in systems biology use to describe their models: Systems Biology Markup Language (SMBL). This is now available in many different packages including Garuda, the software package pushed by the conference organizer Hiroaki Kitano.
The goal is to see how concentrations of reactant species (metabolites, proteins, even spatial parameters such as volume) affect product species. This may even extend to cell sizes, cell populations and gene dynamics.
But perhaps the most exciting potential is the use of such models for personal diagnostics. How does this work? It turns out that the concentration of background reactant species is a unique personal fingerprint. Assuming the network is invariable such as the MAPK/ERK pathway, dynamic modeling can predict the efficacy of different therapeutics for different background reactant species, or different individuals. This has already led to breakthroughs in blood diagnostics.
The Genomic Landscape
It seems silly to say, but it looks like genomics has finally arrived. Yeah, we've had the human genome for a decade. But that's just it - one single human genome, and a weird hybrid one at that. As many commentators in the literature have pointed, knowing a single genome doesn't necessarily help you decipher what it means.
But what has changed is a decade of bloody-minded engineering of DNA sequencing technology which has brought down the sticker price of sequencing a genome to a cheap cheap price of ~$1000. And we mean real genome sequencing, not the opportunistic cherry-picking of 23-and-me.
Many groups, including Leroy Hood's institute, are gearing up for this coming flood of genomes. This flood will mean we can finally measure the real variation in the human genome across actual populations. We can finally shake ourselves of the idea of the single canonical human genome, and study variations - across populations, generations, and even lifetimes. These differences will be key to finding new stuff in the genome. I'd hazard a metaphor and call the the dawn of genomic-based discovery, and hopefully, diagnostics.
Already we're seeing the fruits of studying populations of genomes in other species. Elizabeth Murchison gave a brilliant talk on the Tasmanian-Devil Facial Tumor Disease, a truly frightening cancer that jumps from one Tasmanian Devil to another, shutting off the immune system of the victim in the process. This involved deep analysis of the phylogeny trees generated from the genomes of sick devils sampled all across Tasmania.
Kathyrn Holt described a fascinating epedimelogical study of Shigella dysentry bacteria by correlating the spread of antibiotic resistance of the Shigella bacteria to the phylogenetic tree generated from sampling different populations of Shigella bacteria around the world.
In a sprawling conference like this, you will see a lot of interesting tech. Here's a few that caught my eye.
Antibiotic cellbots is a clever idea for drug delivery with peptides. Peptides, though intrinsically harmless, are hard to administer in our bodies because they'll just be degraded in the stomach. Yiannis Kaznessis described how you might insert the gene for a therapeutic peptide into a bacteria - the cellbot - and use the live bacteria as a drug delivery mechanism as it invades your body.
Tissue modeling. I've done all sorts of modeling including my fair share of "thinking" about multi-scale modeling from molecules to, say, the brain. Peter Hunter points out a level of detail that I've rarely considered, but is absolutely vital - that of tissue modeling. At some point, cells cohere together into a material with physical, chemical and even fluid-dynamics properties. Tissues must modeled properly using engineering techniques, and only then can we really start to model organs.
Serial Block Electron Microsocopy is the technique pioneered and exemplified by Mark Ellisman. It's a subtle combination of slicing, freezing, and electron microscopy that enables the visualization of intact biological tissue at tens of Angstroms resolution. This technique generates breathtaking images, which have resolved many questions about cellular structures - the most famous being the visualization of the tiled intercalation of astrocytes, large connecting neurons. This technique is also a key technology underlying the visualization of the mouse connectome.
Click chemistry, a technology pushed by Leroy Hood, has the potential to displace immuno-precipitation as the method for identifying proteins. The typical method is to inject mice with your protein of choice, and harvest the mouse antibodies that recognize your protein. Surely there is a more elegant way? Though not perfected yet, click chemistry makes that seductive promise. The idea is to model two cyclic peptides that will bind to a protein. And if the two peptides are close enough, spatial connectors will be found in libraries of azide and alynes that will accurately fix the two cyclic peptides in the right orientation. Thus your new protein probe will be generated entirely rationally.
In sum, I heartily recommend the International Conference on Systems Biology. It has a strong focus yet remains broad enough to be of interest to at an interdiscipary audience. More importantly, I've learnt that systems biology leads directly into systems medicine, a gloriously embryonic field - they clearly haven't figured it out yet; they've barely just begun.