Systems biology is a developing scientific field that investigates living organisms as they naturally exist. Unlike traditional methods, it integrates information from multiple disciplines such as biology, physiology, and biochemistry to enhance the understanding of how organisms operate. This integration requires sophisticated and efficient methods for processing and interpreting data. A variety of techniques have been developed to compare biological networks, often utilizing representations derived from graph structures, in which vertices correspond to biological components and edges represent their interactions. This work addresses a computationally hard ( \(\mathcal{N}\mathcal{P}\) -hard) problem involving heterogeneous biological networks, with a particular emphasis on how metabolic processes are influenced by genomic information. A directed graph D is used to represent the metabolic network, whereas an undirected graph G reflects gene proximity, both sharing the same vertex set. We propose deep learning-based methods, particularly leveraging graph neural networks, to identify paths in D such that the corresponding vertices are part of a connected subgraph in G. These paths represent sequences of metabolic reactions driven by enzymes encoded by genes located in close proximity, highlighting biologically meaningful patterns.