GPSVS

Caco-2 permeability prediction, based on an experimental model (Caco-2 cells monolayer) that evaluates the intestinal absorption of drugs [30], is derived from known literature datasets - see [31] for a review. This model was confirmed with GPSVS. 

Water solubility (thermodynamic) prediction, based on various literature datasets [32, 33], is comparable to other models [29]. This model was confirmed with GPSVS. 

ChemGPS-VolSuif (GPSVS) in Clinical PK Property Modeling I 2S7... 

PGA models are usually interpreted by comparing descriptor loadings (their contribution) to the latent variables, which relate to physical meaning. However, in GPSVS, the first two components are directly related to measured properties. 

The GPSVS map in Fig. 11.5 shows WOMBAT-PK drugs color-coded according to oral bioavailability. This plot does not take into account the contribution of active processes (e.g., active efflux, intestinal metabolism, etc.) to this composite parameter. For clarity, an orthogonal view is given in Fig. 11.5. In the absence of individual evidence for each drug, low rank values for G1 and VD s (VD s <... 

Figure11.5. GPSVS plot for 613 drugs with known oral bioavailability. Colors codes correspond to Rank30ral (inset and Table 11.2). Cut-offlines at GPSVl = -6 and GPSVS2 = 5 and the four defined quadrants are shown.

LoV, HiV, LoC, and HiCvalues (and their combinations) are defined in text. Unknown indicates that one or both and Cl values were not available. To the right of each quadrant (Q1-Q4) column, % refers to the percentage for that column, related to the number of drugs in WOMBAT-PK analyzed with GPSVS 154 poor, 283 medium and 176 good %Oral values the total row % values relate to the number (613) of analyzed drugs. 

Having established that GPSVS is compliant with the BCS system [5] regarding oral bioavailability, we further discuss its performance in comparison to RoS compliance [6] for the WOMBAT-PK dataset. RoS compliance for drugs with respect to %Oral, Cl and is shown in Table 11.4. Without examining VD j and Cl, 0 and 1 RoS violations capture 80% and 1S% of the total number of drugs, respectively. 

Figure 11.6. Oral bioavailability distribution comparison betw/een GPSVS quadrants and Ro5 violations for 613 drugs. Color codes correspond to RankSOral (inset and Table 11.2). See text for details.

Ro5 < 2, yet poor oral availability) and 4 are false negatives (Q2, Ro5 > 2, but medi-um/good oral availability). Thus, the combination of GPSVS and Ro5 has a 0.7% false negative rate, and a 5% false positive rate. We conclude that this combination provides a very effective tool for ADME filtering. 

Figure 11.7. GPSVS score accuracy for PENGUINS Scores from VolSurf (x-axis) for the 1000 structures enumerated according to Schemes 11.1 and 11.2 are compared with the scores from PENGUINS (y-axis) The values are 0.91 for GPSVSl scores (a), and 0.9 for G PSVS2 scores (b).

---