Aqueous solubility: the property that quietly kills oral programs

A drug that does not dissolve cannot be absorbed, and a drug that is not absorbed cannot work. Aqueous solubility is the least glamorous number in a candidate’s profile and one of the most common reasons an otherwise excellent oral molecule stalls in development. Potency gets the attention; solubility gets the program cancelled.

Two numbers that are both called “solubility”

Be precise about which solubility you mean, because the two values can differ by orders of magnitude.

• Kinetic solubility is measured by diluting a concentrated DMSO stock into aqueous buffer and watching for precipitation. It is fast, cheap, and runs in high throughput during early screening — but it overestimates real solubility because the compound never reaches its stable crystalline form.
• Thermodynamic solubility is measured by equilibrating solid crystalline material in buffer to saturation. It is the value that actually governs dissolution in the gut. It is slower and needs real solid material, so it arrives later in a program — often after chemists have already fallen in love with a kinetically “soluble” compound.

The BCS map

The Biopharmaceutics Classification System places every oral drug on a two-by-two grid of solubility against intestinal permeability:

• Class I — high solubility, high permeability. The easy case.
• Class II — low solubility, high permeability. Absorption is dissolution-rate-limited; this is where formulation tricks earn their keep.
• Class III — high solubility, low permeability. Dissolves fine, struggles to cross the membrane.
• Class IV — low solubility, low permeability. The danger zone, and where a surprising number of potent kinase inhibitors land.

Butler and Dressman later refined this into the Developability Classification System, which focuses on solubility in the volume of fluid actually available in the small intestine and on the way solubility and permeability can compensate for each other. The practical message of both frameworks is the same: solubility only matters in the context of dose and permeability, not as an abstract number.

Why molecules refuse to dissolve

Poor solubility usually comes from one of two opposite causes, and the fix is different for each:

• “Brick dust” — a tightly packed, high-melting crystal lattice. The molecule is not especially greasy; it simply will not let go of its neighbors. Disrupting planarity, adding a twist or a flexible substituent, lowers the melting point and improves solubility.
• “Grease ball” — high lipophilicity (logP). The molecule would rather sit in lipid than in water. Trimming logP, adding a polar or ionizable group, is the lever here.

This is the tension behind Lipinski’s Rule of Five, whose original paper was literally titled around estimating solubility and permeability. The familiar cutoffs (molecular weight under 500, logP under 5, no more than five hydrogen-bond donors and ten acceptors) are not laws of physics — they are a statistical fence around the region where oral absorption tends to be tractable.

Predicting it in silico

Solubility (usually reported as logS, the log of molar solubility) is one of the more reliably predictable ADMET endpoints because there is a lot of public data. Delaney’s ESOL model showed years ago that a handful of simple descriptors gets you a usable estimate; modern graph neural networks trained on curated datasets do better. The honest caveat is the same as for every QSAR model: predictions degrade for chemistry outside the training distribution, so treat a computed logS as a triage signal, not a measurement.

Try the prediction yourself

Liganx’s ADMET panel runs an admet-ai model ensemble on every compound after a successful dock, and aqueous solubility is one of the properties it returns alongside the cardiac and liver readouts. Open Studio, dock a candidate, then open the ADMET pill on the result row to read its predicted solubility next to potency. Liganx brings molecular docking online into the browser, so running molecular docking and the solubility forecast in the same place lets you catch a brick-dust problem before you commit to the synthesis.

Primary sources

• Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Deliv Rev 46, 3-26 (2001). doi:10.1016/S0169-409X(00)00129-0
• Butler JM, Dressman JB. The developability classification system: application of biopharmaceutics concepts to formulation development. J Pharm Sci 99, 4940-4954 (2010). doi:10.1002/jps.22217
• Delaney JS. ESOL: estimating aqueous solubility directly from molecular structure. J Chem Inf Comput Sci 44, 1000-1005 (2004). doi:10.1021/ci034243x