Geology Club Seminar

Heme-copper oxygen reductases (HCOs) are widely distributed respiratory enzymes playing a pivotal role in aerobic respiration and denitrification. Previously characterized members from this enzyme superfamily are known to catalyze oxygen reduction and nitric oxide reduction. Using comparative molecular phylogenetic analysis, we discovered several new subfamilies, greatly expanding the functional diversity of the HCO superfamily. We hypothesized that several of these new families perform nitric oxide reduction, indicating that the nitric oxide reduction evolved multiple times in this superfamily. Using biochemical techniques, we confirmed that one of these newly discovered subfamilies from a bacterium Rhodothermus marinus, can convert nitric oxide to nitrous oxide.

To further explore the functional adaptation of enzymes from the HCO superfamily, we used deep learning methods, developing a neural network based algorithm to classify protein subfamilies. Distinct from conventional phylogenetic models and Hidden Markov models, this algorithm uses an attention mechanism, applying greater weight to parts of the sequence that it identifies as having features specific to each class when making the classification. This computational tool uncovered additional novel functional motifs in the HCO protein superfamily and is now being applied to other poorly characterized proteins. Given the abundance of genomic sequence data, application of this model to evaluate characteristics of homologous protein families has the potential to provide insight into their functional diversity and evolutionary history.