Water solubility

There are 15 predictive models related to water solubility in ADMET Predictor™. They come in three distinct groups of five models each:

  1. Group based on MHSw-MP (Meylan model of native solubility with melting point as input. Meylan et al; 1996)
  2. Group based on MHSw-NoMP (Meylan model of native solubility without melting point as input. Meylan et al; 1996)
  3. Group based on S+Sw (general Simulations Plus model of native solubility)
In addition, the Simulations Plus S+Sw model is provided in two versions depending on the type of input: 2D models for SMILES, 2D RDF and 2D SDF inputs, and 3D models for 3D RDF and 3D SDF inputs. Perhaps the term "native" warrants some explanation. Imagine a flask containing pure water; no buffers, no salts, no acids/bases, no DMSO, etc. Now imagine a solid chemical compound in its pure, neutral form (i.e., not a salt) dropped into the flask. If there is enough solid compound to reach saturation, then the equilibrium concentration of the compound in question after the flask is shaken for sufficiently long time at room temperature is called the native solubility. If the compound has ionizable functional groups, then naturally its dissolution in water may affect pH due to ionization. The equilibrium pH at saturation is called, as expected, the native pH.

Only the native solubility models in each of the three groups are independent, i.e., based on chemical structures and measured melting point, where applicable. The remaining models in each group are derived from the corresponding native solubility model at its root:

  1. Native pH (pH; calculated from native solubility and ionization constants using chemical equilibrium theory)
  2. Intrinsic solubility (IS; solubility of the neutral species only, calculated from native solubility, native pH, and ionization constants using chemical equilibrium theory)
  3. Salt solubility factor (SF; the ratio of salt solubility to intrinsic solubility; regression models that use the appropriate intrinsic solubilities as descriptors)
  4. Solubility at specified pH (Sp; calculated from intrinsic solubility, salt solubility factors, and ionization constants using chemical equilibrium theory)
The Simulations Plus solubility (S+Sw) model is based on proprietary artificial neural network ensemble (ANNE) created by ADMET Modeler™. The figures below show the correlation between observed and predicted solubilities for 3D model.

S+Sw graph
ADMET Predictor 3D S+Sw Model Validation

In a recent independent test involving 122 drugs, ADMET Predictor solubility models have been shown to have the best predictive performance among all the tested commercially available methods. (Dearden; 2006).

Meylan WM, Howard PH, Boethling RS. "Improved method for estimating water solubility from Octanol/Water partition coefficient." Env Tox And Chem. 1996; 15:100-106.

Dearden JC. "In silico prediction of aqueous solubility." Expert Opin Drug Discov 2006; 1(1): 31-52