Streptococcus pneumoniae is the most common human respiratory pathogen, and β-lactam antibiotics have been employed to treat infections caused by S. pneumoniae for decades.

β-lactam resistance is steadily increasing in pneumococci and is mainly associated with the alteration in penicillin-binding proteins (PBPs) that reduce binding affinity of antibiotics to PBPs. However, the high variability of PBPs in clinical isolates and their mosaic gene structure hamper the predication of resistance level according to the PBP gene sequences.

A research group led by Prof. FENG Jie at Institute of Microbiology of the Chinese Academy of Sciences developed a systematic strategy for applying supervised machine learning (SL) to predict antimicrobial susceptibility testing (AST) of β-lactam antibiotic resistance. The study was published in Briefings in Bioinformatics.

The published PBP sequences with minimum inhibitory concentration (MIC) values and the sequences from NCBI database without MIC values were served as labelled data and unlabeled data, respectively. The performances of SL models were evaluated by cross-validation: the labelled data set was randomly split into 80% training set and 20% test set 100 times.

The researchers demonstrated the association between amino acid loci as an approach for feature reduction using only HVLs or sequence fragments during the building of machine learning models. The cefuroxime and amoxicillin resistance can be predicted well only using fragment from pbp2x (750 bp) and a fragment from pbp2b (750 bp), which allows one Sanger sequencing reaction to predicate the resistance phenotype.

Furthermore, they evaluated the precision of predication model by constructing the mutants containing the pbps from S. pneumoniae strains, of which genomes are available in NCBI database, and their phenotypes were predicated according to the model.

The model was tested by predicting resistance phenotypes of a local clinical strain collection. Both these approaches validated that the SL model could predicate the phenotype accurately. 

Besides, they researchers revealed a correlation between resistant phenotype and serotype and phylogenesis in more than 8000 S. pneumoniae strains available from NCBI database by applying the approach, which facilitates the understanding of the worldwide epidemiology of S. pneumonia.

Overall, this study established an effective genotypic AST approach for detecting β-lactam resistance levels in S. pneumoniae. It demonstrated that the highly variant amino acid loci (HVLs) are associated with antibiotic resistance using the unlabeled data in public database in NCBI.