Machine Learning Approach Identified Multi-Platform Factors for Caries Prediction in Child-Mother Dyads.

Untreated tooth decays affect nearly one third of the world and is the most prevalent disease burden among children. The disease progression of tooth decay is multifactorial and involves a prolonged decrease in pH, resulting in the demineralization of tooth surfaces. Bacterial species that are capable of fermenting carbohydrates contribute to the demineralization process by the production of organic acids. The combined use of machine learning and 16s rRNA sequencing offers the potential to predict tooth decay by identifying the bacterial community that is present in an individual's oral cavity. A few recent studies have demonstrated machine learning predictive modeling using 16s rRNA sequencing of oral samples, but they lack consideration of the multifactorial nature of tooth decay, as well as the role of fungal species within their models. Here, the oral microbiome of mother-child dyads (both healthy and caries-active) was used in combination with demographic-environmental factors and relevant fungal information to create a multifactorial machine learning model based on the LASSO-penalized logistic regression. For the children, not only were several bacterial species found to be caries-associated (Prevotella histicola, Streptococcus mutans, and Rothia muciloginosa) but also Candida detection and lower toothbrushing frequency were also caries-associated. Mothers enrolled in this study had a higher detection of S. mutans and Candida and a higher plaque index. This proof-of-concept study demonstrates the significant impact machine learning could have in prevention and diagnostic advancements for tooth decay, as well as the importance of considering fungal and demographic-environmental factors.

Functional selection of vaccine candidate peptides from Staphylococcus aureus whole-genome expression libraries in vitro.

An in vitro protein selection method, ribosome display, has been applied to comprehensively identify and map the immunologically relevant proteins of the human pathogen Staphylococcus aureus. A library built up from genomic fragments of the virulent S. aureus COL strain (methicillin-resistant S. aureus) allowed us to screen all possible encoded peptides for immunoreactivity. As selective agents, human sera exhibiting a high antibody titer and opsonic activity against S. aureus were used, since these antibodies indicate the in vivo expression and immunoreactivity of the corresponding proteins. Identified clones cluster in distinct regions of 75 genes, most of them classifiable as secreted or surface-localized proteins, including previously identified virulence factors. In addition, 14 putative novel short open reading frames were identified and their immunoreactivity and in vivo mRNA expression were confirmed, underscoring the annotation-independent, true genomic nature of our approach. Evidence is provided that a large fraction of the identified peptides cannot be expressed in an in vivo-based surface display system. Thus, in vitro protein selection, not biased by the context of living entities, allows screening of genomic expression libraries with a large number of different ligands simultaneously. It is a powerful approach for fingerprinting the repertoire of immune reactive proteins serving as target candidates for active and passive vaccination against pathogens.