GNNome is a graph neural network based framework that aims to perform de novo genome assembly. It works by representing a genome as a graph, then assigns probability scores to edges corresponding to if it thinks those two nodes in the genome overlap. However, these graphs are imperfect, and we hypothesize that we can find this missing information by searching all pairwise overlaps between the genome's fragments. In this paper, we create a pipeline to find and introduce this missing information by re-training and re-purposing GNNome's overlap edge prediction capabilities. In doing so, we are able to consistently improve upon GNNome, and outperform other state-of-the-art genome assemblers.