Molecular communications, particularly via extracellular vesicles (EVs), play a critical role in the tumor microenvironment by influencing processes such as anti-tumor signaling and cellular responses like migration. EVs function as signaling pathways mediators, affecting the target cells and initiating feedback loops that establish a bilateral communication system. Evaluating the distance between the two cells remains challenging due to the complexity of these interactions, especially in closed-loop systems. Overcoming this challenge, we train a neural network (NN) using features extracted from raw data, including the numerical differentiation of received molecules and the amplitude and location of resulting peaks. Our major contribution is to provide a plausible explanation of the NN operation. By applying explainable artificial intelligence (XAI) frameworks, such as Shapley values and manual permutation importance, we provide deeper insights and proof of correctness for the NN-based distance estimator. While existing work often relies on local interpretable model-agnostic explanations (LIME) and individual conditional expectation (ICE) for model interpretability, these methods have limitations in capturing the complex feature interactions and nonlinearities inherent in biological systems like molecular communication. Our findings underline the potential of XAI in making complex molecular interactions more transparent, providing critical understanding of the tumor microenvironment, and assisting in the development of more targeted cancer treatments.