191
Part Five · Convergence

Case study: cytochrome P450 and drug metabolism

A handful of liver enzymes, the cytochrome P450 family, metabolize most of the drugs you will ever take; one of them, CYP3A4, handles about half. When two drugs compete for the same P450, or one switches it off, the other can build up to dangerous levels: the source of many drug–drug interactions. P450s do their chemistry with an iron atom held in a heme ring, passing through high-valent iron–oxo states whose spin configurations are delicate. They are a natural test case for quantum computing in pharmacology.

On the plate: where a quantum computer would overtake classical methods for P450 models.

The chemistry and the difficulty

P450 activates molecular oxygen and inserts one oxygen atom into a carbon–hydrogen bond of the drug, making it more water-soluble so the body can excrete it. The key intermediate, “compound I”, an iron(IV)–oxo with a radical spread over the heme ring, has nearly degenerate spin states, and which spin surface the reaction follows (the “two-state reactivity” proposed by Sason Shaik and colleagues) affects rates and products. Different density functionals disagree on spin-state energy gaps by many kcal/mol, which is exactly the regime where multireference methods, and eventually quantum computers, are needed. Established as a hard problem.

The advantage frontier

In 2022 Joshua Goings and colleagues at Google and Boehringer Ingelheim built a ladder of P450 active-space models of increasing size, solved the small ones with classical methods, and estimated quantum resources for the large ones, to locate where a fault-tolerant quantum computer would beat the best classical heuristics. The plate shows a schematic of that comparison: classical cost rising exponentially with active-space size at rate 0.25 decades per orbital, against a quantum cost that grows polynomially but starts high, here 10410^{4} seconds. The crossover moves sharply with assumptions about classical growth, which is why the frontier is uncertain: better classical heuristics push it right. Their conclusion was that quantum advantage for such problems is plausible at active spaces beyond what classical methods can handle reliably, with error-corrected machines of millions of physical qubits. Speculative

The critique

Predicting drug metabolism in practice depends on much more than the energy of compound I: which drug binds, in which orientation, how often, which site is oxidized, and how fast the products leave. Machine-learning models trained on metabolism data already predict likely sites of metabolism reasonably well. A quantum computer could sharpen the electronic-structure step, the spin-state energetics, but its value to drug development depends on the whole workflow, and on whether that step is the bottleneck for a given question. Speculative

What you now have

Why P450s matter for drug interactions, compound I and two-state reactivity, why DFT is unreliable for its spin states, the Goings et al. advantage-frontier study, and why the workflow, not one energy, determines value.

· · ·

Three questions to carry with you.

What does P450 do to drugs?tap
Inserts an oxygen into a C–H bond, making the molecule easier to excrete.
Why are its spin states hard?tap
Nearly degenerate iron–oxo configurations on which density functionals disagree.
What moves the advantage frontier?tap
The growth rate of the best classical methods and the overhead of quantum hardware.

And one to work out. Classical time is 100.25n10^{0.25n} seconds for nn orbitals; quantum time is 10410^4 seconds. At what nn do they cross?

n =

Sources: F. P. Guengerich, Chem. Res. Toxicol. 14, 611 (2001). S. Shaik et al., Chem. Rev. 110, 949 (2010). J. J. Goings et al., Proc. Natl Acad. Sci. USA 119, e2203533119 (2022). The frontier plot is a schematic with adjustable assumptions, not Goings et al.’s data.

Next · Chapter 192
Drug binding and docking: where quantum helps and where classical wins

A 1.36 kcal/mol error changes predicted binding affinity by 10× ✓. Where does that error come from?

Turn the page