Introduction: Computational Methods For Single-particle .

Introduction:Computational methods forsingle-particle cryoelectron microscopyCS/CME/Biophys/BMI 371Feb. 15, 2018Ron Dror1

October’s Nobel Prize in ChemistryAwarded to Jacques Dubochet,Joachim Frank and RichardHenderson and "For developingcryo-electron microscopy for thehigh-resolution structuredetermination of biomolecules insolution"2

Single-particle electron (cryo) microscopy We want the structure of a “particle”: a molecule (e.g., protein) or a well-definedcomplex composed of many molecules (e.g., a ribosome) We spread identical particles out on a film, and image them using an electronmicroscope The images are two-dimensional (2D), and each particle is positioned at a different,unknown angle. Given enough 2D images of particles, we can computationally reconstruct the 3Dshape of the particleElectron beamParticlesImages3Image from Joachim pdf

Improved computational method forreconstructing 3D particle shape Raw electron microscopy (EM) images are very noisy A new software package (Relion) that uses Bayesian statisticsto prevent overfitting to noise has substantially improvedcutting-edge single-particle results4Image from Joachim pdf

0.6 0.6Rosetta0.5 0.5Automated structure refinementFourier shell correlation0.7 0.70.4 0.40.7 0.70.6 0.63.4Å 3.4Å0.5 0.50.4 0.4 Once one has the0.3 moleculeshape (a “density0.30.20.20.2map”), one can modelin the actual atoms0.10.3 0.30.2DepositedDepositedRosettaRosetta0.9 0.90.8 0.8Fourier shell correlation0.80.1 0.10.0manually, and it’s tricky– This is usually done0.00.00 0.05 0.10 0.15 0.20 0.25 0.30 0.350.1 0.2 0.3 0.4 0.5 0.6 0.7 0.80.00 0.05 0.10 0.15 0.20 0.25 0.30 0.350.2 0.3 0.4 0.5 0.6 0.7 0.8Resolution (Å )Deposited– Oneof next week’s paper presentsResolution (Å )an automatedDepositedmethod for improving (refinement) manual taRosetta111RosettaensembleRosettaensemblechain kMitoribosomechain k0.10.15Wang etofal.,theeLife5:e17219(2016) ribosome (EMD-2762) shows improvementsRefinement of the large subunithumanmitochondrialto allMethodssubunits.10:584(A) ScatterplotsofLi et al., Nature(2013)ement of the large subunit of the human mitochondrial ribosome (EMD-2762) shows improvements to all subunits. (A) Scatterplots ofality for each of the 48 protein chains compare the deposited (x-axis) and Rosetta (y-axis) models using MolProbity. On the left, theor each of the 48 protein chains compare the deposited (x-axis) and Rosetta (y-axis) models using MolProbity. On the left, thety scores of all 48 protein chains are compared, where a lower values indicates a better model geometry. On the right, the percentage of

Recovering a conformational ensemblefrom EM images Real biomolecules (and complexes) don’t exist injust a single conformation. They interchangerapidly between different conformations.– Each EM image reflects just one conformation Usually one reconstructs just a single 3Dstructure from a collection of images– Or, perhaps, two or three 3D structures One of next week’s papers aims to recover a fullensemble (that is, the full range of conformations— essentially a “movie”)– Uses manifold embedding methods6

Background information My slides on single-particle electron microscopyfrom CS/CME/Biophys/BMI 279:– re15.pdf My slides on Fourier transforms and convolutionfrom CS/CME/Biophys/BMI 279:– re9.pdf For more detail, see the paper “A Primer to SingleParticle Cryo-Electron Microscopy” (listed on thecourse website as an “additional paper” for nextThursday)7

Introduction: Computational methods for single-particle cryoelectron microscopy CS/CME/Biophys/BMI 371 Feb. 15, 2018 Ron Dror 1. 2 October’s Nobel Prize in Chemistry Awarded to Jacques Dubochet, Joachim Frank and Richard Hende