The organization of metabolic reactions and protein synthesis in eukaryotic cells requires complex machinery that maintains the creation and functionality of specialized compartments and controls the specific subcellular location of the respective proteins 
. The different membrane enclosed compartments (Endoplasmic Reticulum (ER), Golgi stacks, or Endosomes) form a dynamically linked network in which vesicles deliver cargo molecules from donor to target compartments 
The key features of vesicle transport are the accurate selection of only the desired molecules into the vesicles and the transport of the vesicle towards the correct target through the crowded intracellular environment 
. While sorting depends on specific (short range) molecular interactions between the proteins forming a vesicle 
, the navigation through the cell requires a long-range orientation (cf. ) 
. Motor proteins can pull vesicle along cytoskeleton tracks 
. This allows the directed motion towards the target, given that the vesicle happens to run on the right track. Considering the large number of cytoskeleton filaments and furthermore their dynamics, finding the right way through the cell is not a trivial task 
. But also the probability to hit a desired target only by diffusion is rather small. The present work investigated the principal interactions of the transport process and the connecting cytoskeleton structures which guarantee that a vesicle is not lost in space.
Modularity and Interactions of Vesicle Transport.
For a rigorous analysis, the large network can be broken into small units. Each vesicle transport step between two compartments forms such an elementary module as depicted in 
. One module includes vesicle budding at the donor compartment, transport, and the fusion process at the target compartment. During their lifetimes compartments and vesicles can maturate and develop into another compartment, for instance the early into the late endosome 
In principle, each vesicle and compartment is an autonomous entity. The initial state determines the temporal development of its location, internal biochemical conversions, and interactions with other objects in the cell. This especially holds for the key proteins of the vesicle-vesicle interaction, i.e. the fusion process. Vesicle fusion is initiated by a docking and tethering state induced by tethering factors 
. Subsequently the binding of SNARE (Soluble NSF Attachment protein REceptors) proteins connects both membranes and promotes the eventual fusion via a cis-trans-conversion (cf. ). The SNARE proteins can be subdivided into the v-SNAREs in the vesicle membrane and the t-SNAREs at the target compartment 
Accordingly, the v-SNAREs have to be loaded into the vesicle during the budding process as shown in 
. The vesicle itself is created by the polymerization of a coat around it, which forms its shape and selects the cargo molecules via transmembrane domains. This coat consists of a variety of proteins, can be classified as COPI, COPII, or clathrin coat, and shows a modular design 
. The variety of proteins involved in the coat formation and cargo selection on the one hand and the need to simplify this complexity in order to build a full-scale model of the vesicle machinery on the other hand can be accounted for by defining each coat complex as just one generic molecule. We define different coat molecules as subtypes of the coat molecule class, which can vary in their affinity for cargo, SNARE, etc. molecules (just like the real complexes vary in their protein composition). Thus the relevant aspect of the complexity of the coat is preserved in our model. The two-compartment model of Heinrich and Rapoport (2005) likewise uses coat A and B and their different preferences for different compartments, cargo and SNAREs.
This ODE-model of Heinrich and Rapoport 
was already able to generate nonidentical compartments and to facilitate the sorting of molecules. However, it omits the spatial aspects of vesicle transport, tracking only number, size, and state of the compartments - like the models of Gong et al. 
and Brusch and Deutsch 
. The most recent model of Birbaumer and Schweitzer 
covers the spatial aspects with an agent-based simulation, but replaces the cytoskeleton by a potential/force field directing the vesicles and neglecting the molecular details of the budding and fusion machinery. Other models include the spatial aspect using a continuous flow approach to describe the vesicle flux, thus neglecting the discrete properties of individual vesicles 
or only cover sub-problems like budding 
and fusion 
Especially when spatial models are considered, the interactions with motor proteins and cytoskeleton filaments have to be considered for navigating the vesicles through the cellular space. Accordingly we propose to include the affinity of motor proteins to the coats so that they are added to the vesicles during the budding process 
. shows the complete network of interactions between the molecule species that are involved in membrane trafficking.
The aim of the present work is to integrate and condense the present knowledge into a 4D spatio-temporal agent-based model. The virtual three-dimensional cell which is set up in order to model vesicle transport contains cytoskeleton structures and all necessary molecule species to drive the membrane trafficking machinery. The structured and event based approach also preserves the inherent stochasticity equal to the stochastic noise and fluctuations in the real number of vesicles. The limitation of agent/molecular interactions to relevant interactions of the model and the separation into interactions within vesicles and between vesicles in a multi-scale manner still keeps the simulation tractable despite the overall complexity. The introduced modularity of the vesicle transport network further improves the handling and allows an easy scale-up from a simple two-compartment setup towards a model containing all compartments.