Molecular weight, TPSA, donors and acceptors

These are the numbers every filter in Part 5 is built from. The theme of this module is that they are {{o:calculated conventions}}, not measured quantities — and that for two of them the same molecule has more than one correct answer.

Molecular weight is the obvious size measure and it is the one every filter uses. It is also the wrong one for some purposes.

Heavy atom count — the number of non-hydrogen atoms — is a better measure of how much molecule you have, because it does not care whether an atom is a carbon or an iodine. Swapping a hydrogen for an iodine adds 126 to the molecular weight and one to the heavy atom count, and for the purpose of asking "how big is this ligand" the second number is the honest one.

This is why heavy atom count, not molecular weight, is the denominator of ligand efficiency in module 22. A metric of binding energy per unit of molecule should not reward you for using light atoms.

TPSA is the surface area contributed by polar atoms — in the standard definition, nitrogen and oxygen and the hydrogens attached to them. Sulfur and phosphorus are included in some variants and not others.

The word topological matters. TPSA is not measured from a three-dimensional surface. It is computed by adding up a contribution for each polar fragment, from a lookup table, based only on connectivity. That is why it is instant and why it is reproducible — and why it takes no account of whether a polar group is buried inside the molecule or exposed on its surface.

The working thresholds:

• below about 140 Ų for oral absorption;
• below about 90 Ų for penetration into the central nervous system.

Both are soft, both have exceptions, and the CNS figure in particular is a population trend rather than a rule about any individual molecule.

This is the part of the module to remember.

A hydrogen bond donor is a hydrogen attached to an electronegative atom — in practice N-H or O-H. That definition is fairly stable across software, though some count the group and some count the hydrogen, so an NH₂ is one donor or two depending on who you ask.

A hydrogen bond acceptor is where it gets genuinely messy. Two definitions are both in wide use:

• Lipinski's — count every nitrogen and oxygen. Simple, and it is what the Rule of Five was actually written against.
• The chemical definition — count atoms with an available lone pair that can genuinely accept. This excludes amide nitrogens, pyrrole-type nitrogens, and often ether oxygens, which are weak acceptors at best.

These give different answers for most drug molecules, and neither is wrong. They are answers to different questions. What is wrong is applying a threshold written against one definition to a count made with the other.

Molar refractivity is a measure of volume and polarisability combined. It appears in the Ghose filter in module 20 and in older QSAR work, and it correlates strongly with molecular weight — which is why it is used less now than it was. It is worth recognising when you meet it, and it is rarely worth optimising directly.

Polarisability itself — how easily an electron cloud distorts — does matter, because it is what determines the strength of dispersion interactions. Larger, softer atoms are more polarisable, which is why iodine makes better van der Waals contacts than fluorine. Module 15 uses this.

Fsp³ is the fraction of carbons that are sp³ — the proportion of the molecule that is three-dimensional rather than flat.

It became popular after analyses linking higher fsp³ to better developability: more three-dimensional molecules tend to be more soluble, more selective and more likely to survive development. The mechanism is not mysterious. Flat molecules stack, so they crystallise well and dissolve badly; flat molecules also present a less distinctive shape, so they bind more things.

The caution is that fsp³ is a description, not an instruction. Adding sp³ carbons at random makes a molecule greasier and more flexible, both of which are bad. What the analyses support is replacing flat with three-dimensional, not simply adding saturation.

The closing point, and the reason this module exists in the shape it does.

Every number in this module is calculated from a set of rules, not measured from a molecule. TPSA comes from a fragment table. cLogP comes from a fitted model. The acceptor count comes from whichever definition the software author chose. Rotatable bonds, from module 6, are the same story.

Three practical rules follow, and they will save you more trouble than any individual number in this module:

• Never compare descriptors across software without checking that the definitions match.
• Never apply a threshold without knowing which definition it was derived against.
• Quote descriptors to the precision they deserve. A cLogP is worth one decimal place at most, and often not that.