Mataya Duncan, Terrence Sylvester, Emilee Walden, Jenniffer Roa Lozano, Emma Turner, Samuel Duncan, Robert F. Mitchell, Duane D. Mckenna, Rich Adams
Recent advances in machine learning are transforming biological research by offering powerful tools to address complex challenges across the life sciences. In particular, deep learning approaches now enable accurate predictions of protein structure and function, opening new avenues for investigating proteomic diversity in non-model organisms. In this study, we conducted a genomic case study that examines the predicted structure and diversity of odorant receptor (OR) proteins in two species of longhorn beetles (Cerambycidae) with divergent life histories: the highly specialized red milkweed beetle (Tetraopes tetrophthalmus) and the broadly polyphagous Asian longhorned beetle (Anoplophora glabripennis). Using leading predictive algorithms, we inferred the structure of beetle-encoded OR genes, compared confidence scores, and assessed protein diversity across OR families and between the two genomes. Unsupervised clustering applied to pairwise protein comparisons revealed an expected strong correlation between structure and sequence, while supporting the evolutionary classification of previously predicted OR groups and revealing new evidence of previously unrecognized OR subclusters. Notably, we identify specific proteins exhibiting substantial structural divergence despite relatively low sequence divergence with other paralogs, suggesting potential outliers subject to unusual evolutionary processes. These results highlight the utility of statistical learning for uncovering patterns of protein evolution and structural diversity in understudied insect genomes.