Coral reefs are essential for marine habitat and ecosystem conservation. Recent developments have enhanced the analysis of coral reefs using single RNA sequence analysis. However, this method has many impediments such as technical dropouts, sparse data, low capture rates, and biological heterogeneity that persist in coral reefs. These challenges are overcome by adopting an imputation algorithm. The proposed method addresses these issues by generating a gene-to-gene relationship graph. The designed Mutual Information based modified framework initially forms a gene graph followed by applying a deep learning algorithm that uses a hybrid neural network architecture for compensating the imputation dropout events in the single cell RNA-sequence data. It also uses structural information from the gene graph and the learned features from the autoencoder to estimate gene expression levels. Evaluation of imputation was done using K-Means clustering metrics, such as Adjusted Rand Index (ARI), Cluster Accuracy (CA), and Silhouette Score (SC). For the proposed method, these metrics were found to be 0.457, 0.656, and 0.044 respectively, which are very promising. Based on the results, we contend that Single cell RNA sequencing has the potential to enhance our understanding of the pollution effects and guide the development of more effective protection strategies for marine ecosystems.