Comparison of Approaches to Biological Simulation

Masami HAGIYA and Masanori ARITA

Department of Information Science
University of Tokyo
7-3-1 Hongo, Bunkyo-ku, Tokyo 113, Japan
e-mail: (hagiya/arita)@is.s.u-tokyo.ac.jp

Biological Simulation

There can be found a number of researches about constructing a model of a biological system and simulating it on a computer for the purpose of gaining deeper understanding of biological phenomena. The purpose of this talk is to examine the possibility of simulating a biological system with a large number of heterogeneous components like genetic regulatory networks and consider its application to biological researches.

For systems like biological lives, which have a large number of heterogeneous components interacting with one another, it is almost impossible to carry out the precise simulation as in the fields of natural science. Constructing a model that can replay on computer simulation the biological emergence from its primitive components should be considered not as a tool but rather as the ultimate goal of a biological research.

In this talk, we will introduce two researches of biological simulation which we have been doing in recent years under the human genome project led by Ministry of Education. The target of both researches is simulation of genetic regulatory networks, but their methods and purposes are rather different.

One of the researches is that of knowledge-based simulation of the genetic regulatory network of Lambda phage, and the other is that of simulation of a fly embryo with a genetic model based on threshold values.

Knowledge-Based Simulation of Regulatory Action in Lambda Phage

Knowledge-based simulation is not simulation of biological lives themselves but rather simulation of biologists' reasoning about biological lives. So results of knowledge-based simulation may be incorrect. The target of this research is not constructing a huge knowledge-base, but refining knowledge-based models to gain preciseness.

We have developed a knowledge-based, discrete-event simulation system to simulate regulatory action in Lambda phage. The simulation centers on the decision between its two developmental pathways: lytic growth and lysogenic growth in various mutational conditions.

The simulation employs two different abstraction levels: a roughly-abstracted level for the non-critical parts, and a precisely-abstracted level for the critical parts. In the former, our model is discrete-event qualitative simulation on knowledge-based system. In the other, our model is based on quantitative chemical equations. To decide the bifurcation in the critical parts, global behavior of the quantitative result is synthesised using qualitative reasoning.

We have gained the following insights as future works.

Automatic unification of two simulation levels.
The treatment of time. We should have a methodology for automatically splitting time into an appropriate number of phases during simulation.
Full automatic construction of global behavior from the precise level.

Simulation of a Fly Embryo

Biological analysis of Drosophila embryogenesis has provided a model of protein interaction in segment formation. In this research, we developed a software system called SIMFLYGEISHA,SIMFLY.

This system is totally different from the quantitative system using differential equations or from qualitative system using rule-based expert systems. SIMFLY treats both quantitative and qualitative model of protein interaction and aims at reducing the model from a quantitative one to a qualitative one. In the system rules, there are only two relations among proteins: promotion and repression, and the system searches optimal threshold values for the given rules to explain protein expression pattern in segmentation.

We intensionally treated the complex relationship among proteins in a qualitative fashion. The actual system must be more complex than our model, but our purpose was to see how a simple qualitative model can explain complex behaviors of embryogenesis.

Discussions

Two researches we have been doing share a common purpose: to construct a model which can explain global behaviors of a system with a large number of heterogeneous components and which is as simple as possible.

It would be nice if two approaches to biological simulation are unified in the future. We hope that techniques developed in one approach will be of your help.

Acknowledgements

We deeply thank Prof. Minoru Kanehisa at Kyoto University and Prof. Toshihisa Takagi at University of Tokyo for giving us fundamental supports and technical advices for the researches discussed here.

References

[1]Masanori Arita, Masami Hagiya and Tomoki Shiratori: GEISHA System: An Environment for Simulating Protein Interaction, Genome Informatics Workshop V, 1994, pp.80-89.

[2]Masanori Arita, Masami Hagiya and Tomoki Shiratori: SIMFLY: The Simulation of a Fly Embryo, Genome Informatics Workshop V, 1994, pp.230-231.

[3]Tomoaki Shimada, Masami Hagiya, Masanori Arita, Shin-ya Nishizaki and Chew Lim Tan: Knowledge-Based Simulation of Regulatory Action in Lambda Phage, First International IEEE Symposium on Intelligence in Neural and Biological Systems (INBS 95), 1995, to appear.