By John Russell
February 5, 2009 | Using computers to model living systems and to make useful predictions about their behavior is a holy grail pursued by many; yet there are far fewer seeking to go in the reverse direction and infer principles from living systems to improve computer science. Corrado Priami, president and CEO of the Centre for Computational and Systems Biology (CoSBi), a partnership between Microsoft, the Italian Government and University of Trento, is hoping to do both.
Formed just three years ago, CoSBi has grown steadily. Today it has about 25 researchers drawn from many disciplines (physics, chemistry, biology, etc). What’s more, CoSBi has already developed a number of freely available prototype tools and is actively using the scientific community to refine its technology. An early version of the platform has made novel in silico predictions about gene transfer mechanisms which University of Trento researchers have taken into the wet lab to try to validate.
If all goes well, says Priami, they will publish the results on the gene transfer work. More recently, Priami has written a short paper, Algorithmic Systems Biology, An Opportunity for Computer Science, which will be published in Communications of ACM (Association for Computing Machinery) in June 2009. The work sketches many of Priami’s ideas on how computer science can be brought to bear on biology and vice versa. The heart of what computer science can learn from biology, he says, is parallelism and robustness—“computers crash too easily; they’re not naturally tolerant, while living systems are robust and adapt well to environment. Maybe we can make better software using principles from living systems,” he says.
In the paper, he writes, “The convergence between computing and systems biology on a peer-to-peer basis is then a valuable opportunity that can fuel the discovery of solutions to many of the current challenges in both fields, moving towards an algorithmic view of systems biology.”
Commenting on these ideas during our transatlantic phone briefing, Priami notes, “There are different levels of cross fertilization that you can envision between the two areas. The first one is that living systems are much more robust than computer systems are today. They are much more adaptive to their environment than computer systems are today. The first step—and probably of interest to our joint venture with Microsoft—is to understand how we can improve the way in which we produce software to be much more tolerant.” Another lesson, he suggests, might be more energy-conserving approaches to computation.
It’s worth noting another point of emphasis from Priami’s paper before plunging into our live discussion on CoSBi’s work on systems biology. In the paper, Priami picks up the ongoing debate over whether computer science is indeed a science, and contrasts it to mathematics with which computer science is often allied in a subordinate, supportive role. He argues computation is a distinct science, in part because its operations necessarily have a physical reality while mathematics may be theoretical or solvable in the abstract (If I haven’t mangled his argument.) He further argues that living systems share much with computation since they too are information processing (i.e. computation) systems in which real events must occur to do the “computation.” Watch for his paper.
Broadly speaking, CoSBI is trying to develop a new programming language, syntax, and toolset to model and simulate living systems. Asked to distinguish the CoSBi approach from Entelos (top-down, ODE-based, disease models), Genstruct (state-based inference), and Gene Network Sciences (unbiased inferencing from data), Priami offers this:
“We are not using mathematical tools; we are using computer science tools. There are lots of professional reasons for doing that because biological systems are highly parallel; you have thousands of interactions that happen simultaneously. Mathematical modeling is mainly an equation and it is combinatorial in size so when you have to describe all those ways in which systems can touch it is not so suitable to model large systems.”
“What we are trying to do is to make models that are transition-based, not state based. State-based systems, in my understanding, describe the differential expressions or other quantities of variables with state changes and try to condense this into relations and [from these] infer the dynamics of the systems. We’re trying to describe not the state, but the transition from one state to another, which is exactly the way computer scientists describe the behavior of distributed programs, so programs that run on different nodes of a network, exchanging messages. It is a newer way of representing the phenomena. “
“Our approach is much more scalable. We do not have theoretical limitations to the size of the model; only technology limitations in terms of the size of the memory we can use. We aim to make models [that have], say, gene regulatory networks and metabolic networks and what people ordinarily do is look at the networks in isolation. We attempt to have the two together interacting in the same system so that we can understand how they interact together. A long term perspective is to be able to model things like immune system at a molecular level.”
A fuller explanation is available on the CoSBi website at http://www.cosbi.eu/Rpty_Activ.php. So far, CoSBi has released five prototype tools that can be downloaded for free:
Beta Workbench (BetaWB) is a collection of tools based on the programming language BlenX, explicitly designed to represent biological entities and their interactions. BetaWB includes the BetaWB simulator, a stochastic simulator based on an efficient variant of the Gillespie Stochastic Simulation Algorithm (SSA), the BetaWB designer, a graphical editor for developing models and the BetaWB plotter, a tool to analyse the results of a stochastic simulation run. http://www.cosbi.eu/Rpty_Soft_BetaWB.php
Cyto-Sim is a stochastic simulator of biochemical processes in hierarchical compartments. The compartments may be isolated or may communicate via peripheral and integral membrane proteins. The native syntax is designed to be a compact and intuitive way of describing chemical systems. Arbitrary kinetic rate functions are permitted, allowing seamless import and export to SBML. Export to Matlab is also facilitated. http://www.cosbi.eu/Rpty_Soft_CytoSim.php
Kinetics Inference (KInfer) is a tool for estimating rate constants of biochemical network models from concentration data measured, with error, at discrete time points. KInfer is inspired by the maximum likelihood estimation and assumes a discretized version of the law of mass action as rate equation. The discretization of the rate equation makes the evaluation of its expectation value analytically tractable. The probabilistic formulation of KInfer’s algorithm guarantees the noise-robustness and the possibility to extend it to a Bayesian treatment of the parameters. http://www.cosbi.eu/Rpty_Soft_KInfer.php
The Simulations and Networks Analyzer (Snazer) is a modular tool designed to aid the processes of visualizing and manipulating reactive models, as well as to share and interpret time-course data produced by stochastic simulators or by any other means. Snazer upgrades the viewer of Beta Workbench and interfaces its output format. It loads biological networks encoded in SBML as well, and stores graph layouts in standard GraphML files. Moreover, it loads time-course data exported in CSV, compresses and prepares them for remote sharing and statistical processing. http://www.cosbi.eu/Rpty_Soft_Snazer.php
Redi is a reaction-diffusion simulator, built to test new diffusion models and algorithms. Redi simulates biochemical systems at the mesoscopic scale of interaction, employing a space discretized variant of the Gillespie SSA. Diffusion coefficient are not fixed, but computed dynamically in a state-dependent way, inspired to the Maxwell-Stephen model of transport phenomena. http://www.cosbi.eu/Rpty_Soft_Redi.php
Ease of use remains an issue, agrees Priami, and it is on his near-term to-do list: “In the next 18 months I hope to be able to develop an interface to our platform that is really useable by biologists.” He also says he hopes to validate some in silico predictions in the lab and perhaps demonstrate effective crosstalk between large systems that demonstrate the scalability of CoSBi’s approach.
Like most in the modeling community, Priami is confident the tools will mature and he thinks they will be adopted rapidly rather than incrementally, once their value has been decisively demonstrated. Whether that’s two years or five years is anybody’s guess.
“We have achieved a lot. We are trying to make a homogenous platform and also a graphical interface that allows biologists to use our tools by hiding as much as possible the technical details from the user and we are integrating all of these appliances,” Priami says. “Of course they are still prototypes so we need large case studies to validate them and we are investigating in this direction.”
The Microsoft relationship remains solid he says, “So far their main interest is understanding how living systems could lead to better software.” CoSBi researchers visit “Microsoft in Cambridge more or less every month. I meet with people in Redmond two or three times a year around ongoing computer science research at Microsoft that can be useful for our development.”
CoSBi has also attracted number of academic and biopharma partners—GSK and the Salk Institute are two—seeking early access to its technology and providing feedback. Priami says these partners often supply data and specific problems which CoSBi then tackles hoping to demonstrate the power of their approach.
It should be fascinating to watch not only the systems biology tools developed by CoSBi but also to monitor how successful it is in taking ideas and techniques from living systems and incorporating them into computer science.
This article first appeared in Bio-IT World’s Predictive Biomedicine newsletter. Click here for a free subscription.