Fig. 2: Comparisons of AnimalGAN results with QSAR predictions for the test set of all 38 clinical pathology measurements. | Nature Communications

Fig. 2: Comparisons of AnimalGAN results with QSAR predictions for the test set of all 38 clinical pathology measurements.

From: A generative adversarial network model alternative to animal studies for clinical pathology assessment

Fig. 2

a Clinical Chemistry measurements. ALP: alkaline phosphatase; ALT: alanine aminotransferase; AST aspartate aminotransferase, DBIL direct bilirubin, GTP γ-glutamyltranspeptidase, LDH lactate dehydrogenase, TBIL total bilirubin, BUN blood urea nitrogen, Ca calcium, Cl chlorine, CRE creatinine, IP inorganic phosphorus, K potassium, Na sodium, A/G albumin globulin ratio, GLC glucose, PL phospholipid, RALB albumin, TC total cholesterol, TG triglyceride, TP total protein. b Hematology measurements. APTT activated partial thromboplastin time, Bas basophil, Eos eosinophil, Fbg fibrinogen, Hb hemoglobin, Ht hematocrit value, Lym lymphocyte, MCH mean corpuscular hemoglobin, MCHC mean corpuscular hemoglobin concentration, MCV mean corpuscular volume, Mono monocyte, Neu neutrophil, Plat platelet count, PT prothrombin time, RBC red blood cell count, Ret reticulocyte, WBC white blood cell count. For each measurement, the performance of the 12 QSARs was represented in a violin plot while the performance of AnimalGAN was denoted by a golden star. The plot was z-score scaled for an improved visual inspection. AnimalGAN exhibited consistently smaller MSE than what can be achieved with QSARs. The violin plot shows the distribution of the 12 QSARs’ performance, with the width at different points indicating the estimated probability density of the data and the inner boxplot displaying the lower quantile (Q1), median (centerline), upper quantile (Q3) and whiskers extending to the minimum and maximum values within 1.5 times the interquartile range (IQR). All data points in this figure are provided in the Supplementary Data 1.

Back to article page