that address real biological problems. Hence the techniques and prototypes developed and tested on proof-of-con-cept examples must scale smoothly to real-life case studies. Scalability is not a new issue in computing; it was first raised several decades ago when computers started to become connected over dispersed geographical regions and the first high-performance architectures were emerging. ³⁹ Algorithmic systems biology can build on the large set of successfully defined and novel techniques that subsequently were developed—particularly in the areas of programming languages, operating systems, and software-development environments—to address the scalable specification and implementation of large distributed systems.

Consider the exploitation, at user level, of the Internet. The dynamics (evolution and use) of the Internet have no centralized point of control and are based both on the interaction between nodes and on the unpredictable birth and death of new nodes— characteristics similar to the simultaneously active threads of interactions in living systems. Yet although biological processes share many similarities with the dynamics of large computer networks, they still have some unique features. These include self-reproduction of components ( dating back to von Neumann’s self-reproducing automata in computing and to Rosen’s systems, closed under efficient causality, in biology), auto-adaptation to different environments, and self-repair. Therefore it seems natural to check whether the programming and analysis techniques developed for computer networks and their formal theories could shed light on biology when suitably adapted.

Such innovation should be facilitated in the life sciences community by presenting computers as high-throughput tools for quantitatively analyzing information processes and systems—tools that can be made greatly customized through software to work with specific processes or systems. In other words, software may be used to plan and control information experiments to serve a myriad of purposes.

The topic of simulation in particular needs some consideration. Simulation has evolved since the early days of computing into a more quantitative algorith-

algorithmic
systems biology
completely adopts
the main assets
of our computing
discipline:
hierarchical,
systems, and
algorithmic thinking
in modeling,
programming and
innovating.

mic discipline: the rules of interaction between components are used to build programs, as opposed to abstracting overall behavior through equations. The execution of algorithmic simulations relies on deep computing theories, while mathematical simulations are solved with the support of computer programs (where computing is just a service). Ex-

22

ecution of algorithms therefore exhibits emergent behavior produced at system level, through the set of local interactions between components, without the need to specify that behavior from the beginning. This property is crucial to the predictive power of the simulation approach, especially for biological applications. The complex interactions of species, the sensitivity of their interactions (expressed through stochastic parameters), and the localization of the components in a three-dimensional hierarchical space make it impossible to understand the dynamic evolution of a biological system without a computational execution of the models. The algorithms that are executed on top of stochastic engines and governed by the quantities described here are fundamental to discovering new organismic behavior and thus to creating new biological hypotheses.

Algorithmic systems biology can also be easily integrated with bioinformatics. An example that would benefit from such integration is the modeling of the immune system, because the dynamics of an immune response involve a genomic resolution scale in addition to the dimensions of time and space. Inserting genomic sequences of viruses into models is quite easy for an algorithmic modeling approach, but it is extremely difficult in a classic mathematical model, which suffers from generalization

45

because a population of heterogeneous agents is usually abstracted into a single continuous variable. ⁸ Deepening our understanding of the immune system through computing models is fundamental to properly attacking infectious illnesses such as malaria or HIV as well as autoimmune diseases that include rheumatoid arthritis and type I diabetes. Also, computer science can exploit such models to further propel research on artificial immune systems in the field of security. ²³

Design principles of large software systems can help in developing an algo-