Search for: All records

The role of the environment in the emergence and spread of antimicrobial resistance (AMR) is being increasingly recognized, raising questions about the public health risks associated with environmental AMR. Yet, little is known about pathogenicity among resistant bacteria in environmental systems. Existing studies on the association between AMR and virulence are contradictory, as fitness costs and genetic co-occurrence can be opposing influences. Using Escherichia coli isolated from surface waters in eastern North Carolina, we compared virulence gene prevalence between isolates resistant and susceptible to antibiotics. We also compared the prevalence of isolates from sub-watersheds with or without commercial hog operations (CHOs). Isolates that had previously been evaluated for phenotypic AMR were paired by matching isolates resistant to any tested antibiotic with fully susceptible isolates from the same sample date and site, forming 87 pairs. These 174 isolates were evaluated by conventional PCR for seven virulence genes (bfp, fimH, cnf-1, STa (estA), EAST-1 (astA), eae, and hlyA). One gene, fimH, was found in 93.1% of isolates. Excluding fimH, at least one virulence gene was detected in 24.7% of isolates. Significant negative associations were found between resistance to at least one antibiotic and presence of at least one virulence gene, tetracycline resistance and presence of a virulence gene, resistance and STa presence, and tetracycline resistance and STa presence. No significant associations were found between CHO presence and virulence, though some sub-significant associations merit further study. This work builds our understanding of factors controlling AMR dissemination through the environment and potential health risks.

 

Transformer-based models have pushed state of the art in many areas of NLP, but our understanding of what is behind their success is still limited. This paper is the first survey of over 150 studies of the popular BERT model. We review the current state of knowledge about how BERT works, what kind of information it learns and how it is represented, common modifications to its training objectives and architecture, the overparameterization issue, and approaches to compression. We then outline directions for future research. 

Large Transformer-based models were shown to be reducible to a smaller number of self-attention heads and layers. We consider this phenomenon from the perspective of the lottery ticket hypothesis, using both structured and magnitude pruning. For fine-tuned BERT, we show that (a) it is possible to find subnetworks achieving performance that is comparable with that of the full model, and (b) similarly-sized subnetworks sampled from the rest of the model perform worse. Strikingly, with structured pruning even the worst possible subnetworks remain highly trainable, indicating that most pre-trained BERT weights are potentially useful. We also study the “good” subnetworks to see if their success can be attributed to superior linguistic knowledge, but find them unstable, and not explained by meaningful self-attention patterns.