Tuesday, February 26, 2019

If computational chemistry methods like density functional theory work well for organic chemistry, why are experimental techniques like X-ray diffraction needed to predict structure, & why isn't drug discovery using computer simulation easy?

TLDR; Highly accurate computational chemistry methods do work very well for organic chemistry - as long as the molecules are reasonably small, isolated and cold. Protein-ligand-systems under physiological conditions are large, dissolved (and perhaps “crowded”) and hot.
Quantum chemistry exhibits bad computational scaling properties, which makes calculations on condensed matter systems like e.g. a protein with a drug molecule in its binding site and dissolved in water (such systems have something like 100000 atoms) at ambient temperature extremely expensive. This is slowly changing, but we are currently barely at a point where the much cheaper molecular-mechanical forcefield calculations are fast enough to calculate the free energy of binding (the measure of binding strength) between a protein and a potential drug molecule on a timescale and with costs palatable to industrial researchers (but those are not particularly accurate…). The reason is that unless your system is ultra-cold, the atoms are moving all the time and you need not a calculation on a single structure but lots and lots of them to get a meaningfull “ensemble” of structures that truly sample and/or represent the many degrees of freedom of movement such a system has.
Structure prediction for organic crystals (small molecules) is more tractable. However, this is still a global optimization problem that again requires thousands of calculations. Nevertheless it nowadays seems to work reasonably well, see e.g. the Report on the sixth blind test of organic crystal structure prediction methods. Now if you again mean predicting the structure of protein-ligand-systems, this is again currently out of reach due to the computational expense.

No comments:

Post a Comment