A graph is a mathematical model representing binary relationships between elements of a set. It is composed of two sets: the set of the elements called the vertices and a set of pairs of vertices called the edges. Graphs appear in many fields from computer networks to chemistry and even scheduling. Often, problems about graphs consist in finding the best possible structure to reduce costs or optimize performance. The best structure is the one that maximizes or minimizes some measure called a graph invariant. These problems are usually solved using heuristics that will try to find a good enough solution because finding the best solution is time-consuming. Extremal graph theory, instead of using heuristics, uses formal proofs to show that specific structures, called extremal graphs, are indeed the ones that reach the extremum value for an invariant given some constraints. The process of finding those extremal graphs can benefit from computer software that produce candidates in order to help researchers. Those candidates then need to be studied, and a formal proof has to be written to show that they are indeed extremal graphs. However, there are no software that aim to help researchers with the writing of a proof. In this work, we focus on a type of proof called the proof by transformation. It consists in showing that, any graph that is not an extremal graph can be transformed, using a chosen set of transformations, into a graph whose invariant value is closer to that of the extremum. We study the possibility of using computers to generate transformations for huge amount of small graphs in order to provide insight about these transformations and their impact on graph invariants. This method, combined with theoretical and technical optimizations, led to the implementation of a tool that has been successfully used in actual research, helping in the writing of proofs by transformation for actual problems in extremal graph theory as well as producing conjectures about the impact of some transformations on the value of graph invariants.