Biopesticides are widely available insect control applications derived from plant, animal, or bacterial proteins. They do not leave harmful residues and are more target-specific than chemical pesticides, but long-term use has led to resistance. Insecticidal protein genes (IPGs) are frequently found encoded in the genomes of arthropod pathogens, especially in the large plasmids found in soil bacteria. However, there are often several similar IPGs found on the same plasmid, which fragments their assembly. Further complicating the search, existing prediction tools analyze one contig at a time, and many IPGs are spread across multiple contigs, but the structure of the genome assembly graph can be used to combine multiple contigs. A new tool, ORFograph, uses this ‘graph-aware’ technique to predict IPGs. Benchmarking ORFograph on genomic and metagenomic datasets yielded both known IPGs that were “hidden” in assembly graphs and potential novel IPGs that had evaded existing tools. This shows that graph-aware gene prediction tools can be used to discover a greater diversity of potential IPGs for phenotypic testing. These results only demonstrate ORFograph's utility to find IPGs, but this pipeline could be generalized to any class of genes.