Fig. 4: Multiple-SP model nanomotion AST performance relative to MIC strips. | Nature Communications

Fig. 4: Multiple-SP model nanomotion AST performance relative to MIC strips.

From: Accurate and rapid antibiotic susceptibility testing using a machine learning-assisted nanomotion technology platform

Fig. 4

a Left: To develop a model for CRO, the MIC distribution of 160 E. coli and K. pneumoniae isolates that were analysed in 4-h nanomotion recordings at 32 µg/ml CRO (Conc., dashed line). Right: Classification according to nanomotion and a 4-SP model, with MIC strip serving as a reference. Circles represent median scores for a PBC samples measured ≥1 time (487 PBC samples and 1485 recordings, Supplementary Data 2). Score is defined as S > 0 > R. Closed circles show correctly classified measurements (TP, TN), open circles show falsely classified measurements (FP, FN). b Left: MIC distribution of 127 E. coli strains at 8 µg/ml CIP (conc. dashed line) analysed to develop a CIP model (210 PBC samples, 573 recordings). Right: Classification according to nanomotion and a 4-SP model, with MIC strip serving as a reference. c Independent testing of the CRO model in (a) on 91 E. coli and K. pneumoniae isolates (left, MIC distribution) and classification (right), with MIC strip serving as a reference. d Independent testing of the CIP model in (b) on 65 E. coli strains (left, MIC distribution) and classification (right), with MIC strip serving as a reference. e Training and test performance for each of the two classification models including results from the PHENOTECH-1 study reporting accuracy [(TP +  TN)/n], sensitivity [TP/(TP + FN)], and specificity [TN/((TN + FP)], with Kirby-Bauer serving as a reference. Data are available in the source data file. Single SP values and scores are available in Supplementary Data 2.

Back to article page