PDF 6up, 4.9M
advertisement
Single Particle Reconstruction with EMAN Methods for High Resolution Refinement in Single Particle Processing Steve Ludtke Asst. Professor, Biochemistry, BCM Co-director,NCMI NCRR GroEL Donghua Chen Joanita Jakana Wah Chiu Jiu-Li Song (UT-SW Med) David Chuang (UT-SW Med) EMAN: http://ncmi.bcm.tmc.edu/eman GroEL 2000 (15 Å) 5000 particles, JEOL 4000 GroEL 2003 (6 Å) 30,000 particles, JEOL 2010F GroEL 2001 (11.5 Å) 5000 particles, JEOL 4000 2005 Animation Unavailable in PDF version Jeol 3000 7 Days of imaging, 910 micrographs 1.06 Å/pix, Nikon 9000 scanner 135 used, 34,868 particles Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Animation Unavailable in PDF version Ca2+ Release Channel Irina Serysheva Wah Chiu Susan Hamilton Ca2+ Release Channel Myofibril 200 kV image of ice-embedded RyR1 (no continuous CF) Plasma membrane T-tubule Terminal cisternae of SR • SR membrane, triggered by DHPR in T-tubule • Homotetramer • ~2200 kDa Tubules of sarcoplasmic reticulum • Releases Ca++ which initiates cross-bridge cycle 1.2 µm 500 Å ~30 Å Resolution Animation Unavailable in PDF version 20 Å Resolution 14 Å Resolution 9.6 Å Resolution Animation Unavailable in PDF version Animation Unavailable in PDF version 270 Å T-tubule membrane cytoplasmic face Cytoplasm 190 Å } TM SR membrane SR lumen SR lumenal face Sequence assignment of observed helices Sequence assignment of observed helices Filter RyR1: Hinge Filter 4864 –NKSEDEDEPDMKCDDMMTCYLFHMYVGVRAGGGIGDEIEDPAGDEYELYRVVFDITFFFFVIVILLAIIQGLIIDAFGELRDQQEQVKEDMETK- 4957 M9 RyR1: M10 Helix 2 Hinge 4864 –NKSEDEDEPDMKCDDMMTCYLFHMYVGVRAGGGIGDEIEDPAGDEYELYRVVFDITFFFFVIVILLAIIQGLIIDAFGELRDQQEQVKEDMETK- 4957 M9 M10 Helix 1 Filter KcsA: 36 -QLITYPRALWWSVETATTVGYGDLYPVTLWGRCVAVVVMVAGITSFGLVTAALATWFVGREQ -119 Pore helix Inner helix Filter MthK: lumenal side (‘out’) Hinge 45 -SWTVSLYWTFVTIATVGYGDYSPSTPLGMYFTVTLIVLGIGTFAVAVERLLEFLINREQ- 103 Pore helix Inner helix Filter Pore helix Filter Helix 2 Helix 1 kink kink Pore helix Inner helix Inner helix MthK RyR1 RyR1/KcsA/MthK KcsA CCD vs. Film CCD Film + Scanner Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Final 3D Model Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Final 3D Model Final 3D Model Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Final 3D Model Final 3D Model Refine from Gaussian Ellipsoid Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Final 3D Model Iteration 1 Iteration 2 Iteration 3 Iteration 4 Iteration 5 How do we get to Higher Resolutions? • Get a better microscope • Find a better microscopist • Algorithm Improvements M(s) = F(s) C(s)2 E(s)2 + N(s)2 C(s) E(s) + N(s) N(s) M(s)2 F(s) C(s)2 E(s)2 + N(s)2 Image Classification ? ? ? Alignment/Registration Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Final 3D Model Alignment/Registration Alignment/Registration Alignment/Registration Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Measures of Similarity • • • • • • Correlation Coefficient Variance (transformed density) Variance (matched filter) Phase Residual Mutual Information etc. ( - )2 = Final 3D Model ( - )2 = ( - )2 = And the Answer is… ( - • • • • )2 = Wiener filter particle Filter reference to match Normalize reference density to particle Calculate variance Model Bias Base Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Classify Particles Align and Average Classes Noisy Align to Final 3D Model 25 100 250 1000 2000 Model Bias Base 25 Noisy (~10% contrast) 100 250 Model Bias Align to 1000 2000 Base 25 Noisy (~10% contrast) 100 Model Bias Base Noisy 250 Align to 1000 2000 Model Bias Align to Base Noisy Align to Iter x4 25 100 250 1000 2000 25 100 Model Bias Base Noisy 250 1000 2000 Model Bias Align to Base Noisy (~10% contrast) Align to Iter x8 25 100 250 1000 2000 25 100 250 1000 2000 Model Bias Base Noisy Align to Initial 3D Model Uniform Projections Build New 3-D Model Particle Images Multi-Classify Particles Align and Average Classes Iter x4 25 100 250 1000 Final 3D Model 2000 The Future • Each particle -> best n classes • More restrictive exclusion from class-avg • • • • • • Better similarity criteria Improved CTF model Per-particle CTF (at least defocus) Beam tilt Better 3-D reconstruction New refinement methodologies
