BIOINFORMATICS Vol. 00 no. 00 2004 Pages 1–8 A knowledge based approach for representing and reasoning about signaling networks C. Baral 1 , K. Chancellor 1 , N. Tran 1 , N.L. Tran 2 , A. Joy 2 and M. Berens 2 1 Department of Computer Science and Engineering, Ira A. Fulton School of Engineering, Arizona State University, Tempe, AZ 85281, USA 2 Translational Genomics Research Institute, 400 N. Fifth Street, Suite 1600, Phoenix, AZ 85004, USA ABSTRACT Motivation: In this paper we propose to use recent develop- ments in knowledge representation languages and reasoning methodologies for representing and reasoning about signaling networks. Our approach is different from most other qualitative systems biology approaches in that it is based on reasoning (or inferencing) rather than simulation. Some of the advan- tages of our approach are: we can use recent advances in reasoning with incomplete and partial information to deal with gaps in signal network knowledge; and can perform various kinds of reasoning such as planning, hypothetical reasoning and explaining observations. Results: Using our approach we have developed the system BioSigNet-RR for representation and reasoning about signa- ling networks. We use a NFκB related signaling pathway to illustrate the kinds of reasoning and representation that our system can currently do. Availability: The system is available on the Web at http://www.public.asu.edu/ ∼ cbaral/biosignet. Contact: baral@asu.edu 1 INTRODUCTION Our goal in this paper is to make progress towards develo- ping a system (and the necessary representation language and reasoning algorithms) that can be used to represent signal net- works and pathways associated with cells and reason with them. We are interested in qualitative representation and rea- soning, which can emulate the reasoning done by biologists (or biochemists) when they analyze and navigate a signaling pathway, but at a much larger scale and with respect to lar- ger networks. This is in contrast to the approaches where signalling and transformation between various compounds are expressed using differential equations (O.Voit, 2000; Mishra, 2002; Antoniotti et al., 2003; de Jong, 2002). In recent years several qualitative approaches have been pro- posed in systems biology (Priami, 2003). Most of these approaches are concerned with modelling and analysis of the model is done via simulation (Peleg et al., 2002; Regev et al., 2001) and perturbation. While certain questions about cell behavior can be easily answered using such an approach (such as the impact of a particular event), it is computationally expensive to answer questions about explaining a particu- lar observation, or planning to alter the cell behavior in a particular way using simulation. In both these cases a simu- lation based approach would entail doing a large number of simulations to find the right explanation or the right plan. Our approach in this paper is a knowledge based approach. We consider the cellular signal network as a knowledge base which can be asked many different kinds of queries. For the various kinds of queries the knowledge base is augmented with various reasoning mechanisms that allow the answering of the queries. This approach is more general than answering standard database queries with respect to a signal network database (Karp et al., 2000; Ogata et al., 1999), as the later is limited to answering queries that can be expressed using a database query language. For example, no existing database query language can express a query whose answer is a plan or an explanation. An important dimension of our approach is that it allows for reasoning mechanisms that gracefully handle incomplete or partial information. This is extremely important as existing signal networks and pathways often have missing or suspec- ted interaction links, or proven interactions whose outputs are uncertain, for example the yeast 2-hybrid interactions mentio- ned by Sambrano (2003). Besides being able to handle such incomplete information, our approach also allows for easy updating (referred to as ‘elaboration tolerance’) of the know- ledge base when new knowledge becomes available. This avoids significant overhauling of the old model or scrapping of the old model and making a new model from scratch. Note that simple frame based approaches (Karp et al., 2000), which are a sub-class of classical logic, are monotonic and hence not elaboration tolerant. © Oxford University Press 2004. 1