CVC Seminar: Wed Apr 6 1:30-3pm ACES 4.304

Radhakrishna Bettadapura

Flexible fitting of subunit assemblies into cryo-EM density maps

In flexible fitting, the crystal structure P of a protein is deformed to maximize a pre-determined correlation measure with a cryo-EM density map M of the same protein. All flexible fitting algorithms ask, and answer, three questions: 1) How can P be deformed? 2) How can P be correlated with M? and, 3) How can deformation and correlation be a) sped up and b) made more accurate? Whereas approaches to flexible fitting in the past have answered questions 1 and 2 fairly well, 3 has been essentially ignored; for instance, most flexible fitting approaches do not quantify either the complexity of their technique or the loss in accuracy due to its application on the PDB. In addition, the question of correlating the PDB with the density map has been answered in nearly identical ways: by blurring the PDB and computing its spatial correlation with the density map.

In this talk, we present a new answer, based on motion assignment to domains, to question 1; a new answer, based on skeletonization of density maps, to question 2; and a new answer, based on rotational and translational FFT speedups, to question 3. We thus present a new flexible fitting algorithm, highlighting the ways in which it is different from older, more tested approaches. We also outline the reasons why we expect it to work.

Notes: The above title and abstract belong to my proposal talk, which I hope to give in May. Wednesday’s talk, while featuring most of the techniques promised above, will not be complete. In particular, it will discuss issues of speed but ignore for the time being those of accuracy; it will contain rather more technical detail than a proposal talk can, or should, tolerate; it will have a low to middling number of pictures; and it will introduce the audience to concepts with which they are already familiar. (The reason for the last is actually quite important: my proposal committee is made largely of professors of mechanical engineering, who do not know about PDBs and alpha helices and Gaussian blurring. )