Integrating diverse biological features such as formativeness of topological properties and gene expression presents a significant challenge due to the complexity of determining each feature's individual contribution to predictive models. Ensuring the outcomes reflect the underlying biological structure of the information within the network, while noise and irrelevant data are at a minimum, is crucial. This study identifies the importance of rigorous pre-analyses to determine statistically significant correlations and joint effects among the pre-process features before the application of machine learning techniques. Using multidimensional datasets, we propose a systematic multi-feature framework that combines optimization graph clustering, weighted Jaccard Similarity, and PCA-based dimensionality reduction to uncover previously uncharacterized gene associations in complex biological systems. The network-based framework has employed graph clustering internally for the efficacious allotment of research resource by identification of genes with significant changes in certain community contexts. Moreover, it provides more in-depth insights into how genes interact in their communities, highlighting patterns and relationships that may be obscured in holistic data analyses. The effectiveness of the proposed approach was validated using the benchmark dataset from the DREAM5 challenge project, illustrating its power in analyzing complex biological networks.