Avoiding Ice IX
I never knew the late author Kurt Vonnegut would best explain what I do for work. Not in the way you’d expect, like with a prosey metaphor for capitalism or a theory of personal fulfillment. Rather, in his ’63 novel Cat’s Cradle, he concisely (and literally) describes the mystery and peril of crystal polymorphism better than I possibly could. I’m not joking, here’s the passage:
There are several ways… in which certain liquids can crystallize — can freeze — several ways in which their atoms can stack and lock in an orderly, rigid way… think of the several ways in which cannonballs might be stacked on a courthouse lawn, of the several ways in which oranges might be packed into a crate. So it is with atoms in crystals, too; and two different crystals of the same substance can have quite different physical properties.
He goes on to describe a crystalline form of water (i.e. ice) called Ice IX, which is so stable that if it contacts liquid water it induces the water to instantly freeze into Ice IX. Even if you find new liquid water somewhere, microscopic Ice IX will freeze it. We call this very real behavior "crystal seeding," but this particular apocalyptic form is indeed fictional.†
When I read this passage as a teenager, the concept of polymorphism (although he never uses this term) reeked so strongly of plot device that I assumed it was wholly fabricated. It wasn't until I was in graduate school that I heard of the crystallographer's boogeyman: ritonavir, an HIV drug which was accidentally seeded by its own "Ice IX," an overly-stable second form of the drug that was not bioavailable and therefore could not treat HIV. The drug was recalled and any attempt to produce the original crystal form would invariably fail and produce the useless form.
The existence of the boogeyman, situations like ritonavir Form II, can actually be predicted. You don't have to take the drug to market and risk lives and your bottom line to figure out if there's another form. At Lavo we simulate millions of theoretical forms and determine if the one you want to put into a pill is the most stable of them. If so, we're pretty confident you've avoided this problem with seeding and indefinitely losing your effective form (a problem we call "disappearing polymorphs"). If we predict the existence of more stable forms, it's usually best to keep doing crystallization experiments to find the missing forms, or to find alternative formulation strategies that don't require the current form.
As the size of molecules grows, so do the number of possible ways in which it can pack. Water is small enough that Linus Pauling essentially just guessed its "main" crystal structure in 1935. The packing of small, rigid drug molecules has only become routine in the last decade. Unfortunately, most drug molecules aren't so small or rigid, and predicting their crystal structures is still considered an open challenge in science. That said, by using AI to make our simulations more efficient and running those simulations in parallel on the cloud, we've been able to predict the crystal structures of surprisingly complex modern drugs, as well as their stability with respect to temperature. We do this often enough, fast enough, and successfully enough that I'm prepared to call it "routine." ‡
So that's what we do. We help ensure drug developers don't get an Ice IX, a ritonavir form II. If they might, we can foresee it and work around it before it finds itself in the medicine cabinets of patients who need the right crystal form to live.
† Water is a highly polymorphic molecule and Ice IX is actually one of the forms of ice, but it doesn't have this deadly trait. Ice "h" is the form you and I are most familiar with — the "h" stands for hexagonal, and this form is responsible for the shape of snowflakes.
‡ I'm describing one of many extremely difficult problems in predicting the solid form behavior of drugs, and materials in general. We work on many related problems; we haven't solved all of them, but we believe routine crystal structure prediction is a crucial step in making drugs "computable."