Core Concepts
This paper introduces a novel method for the automated construction of minimal graph transformation models from input transition systems, capturing the dynamics of a system. The method can perform both lossless and lossy compression, the latter suggesting completion of the input dynamics.
Abstract
The paper introduces a formal method for compressing the known transitions (explicit semantics) of a graph transformation model into a set of rules (implicit semantics). The method has two main working modes:
Lossless compression: The constructed rule set exactly reproduces the input transition system.
Lossy compression: The constructed rule set over-approximates the input transition system, introducing new transitions not present in the original data.
The key aspects of the method are:
Defining the notion of a generating rule set, which can reproduce the behavior captured in an input transition system.
Identifying the maximum rule for each transition, which serves as the upper bound for the minimal generating rule set.
Exploring the candidate rule space as the downward closure of the maximum rules under the subrule relation.
Formulating the problem of finding a minimal generating rule set as an instance of the set cover problem, which can be solved using integer linear programming.
The paper discusses various use cases for the method, including:
Reverse engineering graph transformation models from empirical data
Complexity analysis of graph transformation models using the size of the minimal generating rule set
Suggesting completion of the input dynamics through lossy compression
The method is illustrated with examples across different graph transformation models.
Stats
The paper does not contain any explicit numerical data or statistics. The focus is on the formal
definition and description of the proposed method.
Quotes
"The explosion of data available in life sciences is fueling an increasing demand for expressive models and computational methods."
"Graph transformation is a powerful formalism well-founded in theory [1, 2, 3]. Graph transformation is not only a universal model of computation, but also enjoys a diverse and growing number of applications across numerous fields and areas, such as software engineering [4, 5], biology [6, 7, 8, 9] or chemistry [10, 11, 12, 13]."
"Identifying this underlying model is a key question in network analysis. The simpler the model explaining the empirical data, the better, being a good rule of the thumb."