Article Figures & Data
Figures
- Fig. 1.
Overview of human VD,ss prediction methods and input parameters (in silico and in vitro data) evaluated in this study.
- Fig. 2.
Summary of model performance of in silico VD,ss prediction methodologies: (A) ATOM in silico set (n = 956 compounds) and (B) ATOM experimental set (n = 254 compounds).
- Fig. 3.
Predicted vs. observed VD,ss using direct ML models: (A) the ML model built was using a smaller data set (287 compounds), and predictions were tested on a large in silico set (956 compounds) and (B) vice versa. Crosslines indicate 2-, 3-, and 10-fold limits.
- Fig. 4.
Predictive performance of mechanistic Kp prediction methods using various combinations of experimental (Exp.) data.
- Fig. 5.
(A) Scatter plot [colored by ionic state reported in Lombardo et al. (2018)]. (B) Kernel density plot showing correlation between observed and predicted VD,ss for 254 compounds using experimental (exp) fup and BPR data as input parameters. Crosslines indicate 2-, 3-, and 10-fold limits.
- Fig. 6.
Predictive performance using adipocyte (Kp fat) and myocyte (Kp muscle) partitioning experimental (exp) data.
Tables
Mechanisms Poulin Homogeneous Berezhkovskiy Rogers and Rowland Lukacova Trapp Intracellular Albumin binding Yes Yes Yes Yes No Neutral phospholipid binding Yes Yes Yes Yes No Neutral lipid binding Yes Yes Yes Yes No Acidic phospholipid binding No No Yes Yes No Cytosolic ion partitioning No No Yes Yes Yes Lysosomal ion trapping No No No No Yes Mitochondria ion partitioning No No No No Yes Membrane potential No No No No Yes Intracellular water Yes Yes Yes Yes Yes Extracellular water Yes Yes Yes Yes No Tissue-specific composition Yes Yes Yes Yes No - TABLE 2
Summary of model performance of in silico VD,ss prediction methodologies for Lombardo intravenous dosing drug set (n = 1352 drugs) divided into two subsets: 1) ATOM in silico set (>940 compounds) and 2) ATOM experimental set (n > 280 compounds)
Method Description Input Parameters n Within 2-Fold Within 3-Fold Within 10-Fold r2 AAFE % ADMET mechanistic Log D, fup, BPR (predicted using ADMET Predictor models) 956 47 65 90 0.25 2.8 287 51 71 92 0.35 2.4 ATOM mechanistic Log D, fup, BPR (predicted using ATOM ML models) 936 45 62 87 0.23 3.1 285 50 66 92 0.38 2.7 Allometry (rat and dog) Rat VD,ss, dog VD,ss (predicted using ATOM ML models) 956 45 66 93 0.28 2.7 283 45 69 96 0.37 2.5 Allometry (rat) Rat VD,ss, rat fup, fup (human) (predicted using ATOM ML models) 956 30 47 79 0.02 4.4 283 23 40 76 0.0 4.9 Allometry (dog) Dog VD,ss, dog fup, fup (human) (predicted using ATOM ML models) 956 28 43 75 0.05 4.9 283 23 38 77 0.12 4.7 Direct ML model for human VD,ss SMILES/MOE descriptors 956 36 55 88 0.14 3.3 285 58 75 98 0.52 2.2 - TABLE 3
Summary of mechanistic VD,ss predictive performance using experimental data (fup, BPR, and log D) as input parameters
Method Description Input Parameters n Within 2-Fold Within 3-Fold Within 10-Fold r2 AAFE % Experimental fup Experimental fup. Other input parameters were predicted by ATOM ML models 254 56 73 93 0.42 2.3 Experimental BPR Experimental BPR. Other input parameters were predicted by ATOM ML models 254 57 79 96 0.51 2.1 Experimental log D Experimental log D. Other input parameters were predicted by ATOM ML models 254 47 70 89 0.29 2.7 Experimental fup, BPR Experimental fup and BPR. Other input parameters were predicted by ATOM ML models 254 65 81 96 0.58 2.0 Experimental fup, BPR, log D Experimental fup, BPR, and log D 254 59 76 93 0.46 2.2 Experimental fup (BPR = 1) Experimental fup, and BPR is assumed equal to 1. Other input parameters were predicted by ATOM ML models 254 63 76 94 0.48 2.1 - TABLE 4
Performance of VD,ss prediction methods utilizing adipocyte and myocyte Kp experimental data
Method Description Input Parameters n Within 2-Fold Within 3-Fold Within 10-Fold r2 AAFE % Kp fat only Kp adipocyte, fup and BPR 189 54 68 97 0.42 2.4 Kp muscle only Kp myocyte, fup and BPR 189 41 65 94 0.43 2.6 Kp fat and muscle Kp adipocyte, Kp myocyte, fup and BPR 189 36 63 92 0.46 2.9 Kp average of fat and muscle Kp adipocyte, Kp myocyte, fup and BPR 189 31 46 83 0.46 3.9 Kp fat and Kp muscle with mechanistic predicted Kp values for other tissues Kp adipocyte, Kp myocyte, fup and BPR, predicted Kp values using ATOM mechanistic models for tissues other than fat and muscle 189 33 56 93 0.48 3.1 Kp fat and Kp muscle with mechanistic predicted Kp values for other tissues using experimental log D Kp adipocyte, Kp myocyte, fup and BPR, predicted Kp values using ATOM mechanistic models for tissues other than fat and muscle 169 27 46 85 0.41 3.9 Lukacova with experimental fup, BPR Experimental fup and BPR. Other input parameters were predicted by ATOM ML models 189 60 76 95 0.50 2.2
Data Supplement
- Supplemental Data -
Supplemental Table S1 - Machine learning (ML) models used to predict input parameters for various in silico VD,ss prediction methodologies
Supplemental Table S2 - Experimental measurement of physicochemical properties and observed VD,ss values (Lombardo et al., 2018) for 254 compounds
Supplemental Table S3 - Experimental determination of adipocyte and myocyte cell partitioning of 189 clinical compounds (Lombardo et al., 2018).
Supplemental Figure S1 - UMAP projection of datasets (a) ATOM in silico and (b)experimental set based on ECFP Tanimoto distances.
Supplemental Figure S2 - Scatter/KDE plots of observed versus predicted using various silico VD,ss prediction methodologies listed in Table 2 (a) ATOM in silico set (>940 compounds) and (b) ATOM experimental set (n>280 compounds). Crosslines indicate 2, 3 and 10-fold limits.
Supplemental Figure S3 - Measured ChromLogD compared to predicted LogD values by (a) ADMET Predictor (b) ATOM ML model.
Supplemental Figure S4 - Scatter/KDE plots of mechanistic VD,ss predictions using various combinations of experimental data (fup, BPR and LogD) as input parameters.
Supplemental Figure S5 - Scatter/KDE plots of mechanistic VD,ss predictions using various combinations of experimental data (Kp fat and Kp muscle) as input parameters.
Supplemental Figure S6 - Scatter plot of mechanistic VD,ss predictions color coded by ionization using various combinations of experimental data (fup, BPR and LogD) as input parameters.
Supplemental Figure S7 - Scatter plots of predicted in silico vs observed VD,ss using (a) Allometry (Rat and Dog) and (b) ADMET Mechanistic models. Crosslines indicate 2, 3 and 10-fold limits.
- Supplemental Data -
In this issue
In Silico Prediction of Volume of Distribution in Humans
