Recent developments in Crank. Leiden University, The Netherlands

Transcription

1 Recent developments in Crank Navraj js. Pannu Leiden University, The Netherlands

2 Current developers Pavol Skubak Ruben Zubac Irakli Sikharulidze Jan Pieter Abrahams RAG de Graaff Willem-Jan Waterreus

3 Substructure factor amplitudes (F A ) Flow of CRANK AFRO, SHELXC CRUNCH2, SHELXD Substructure coordinates Phases, phase probabilities Density modified phases Protein structure BP3, SHELXE SOLOMON, DM, SHELXE, RESOLVE, PARROT ARP/wARP, RESOLVE, BUCCANEER

4 CRUNCH2: A program for substructure detection. Algebraic approach based on rank reduction of Karle/Hauptman matrices. Considers a higher order collection of reflections over triplets/tangent formula. de Graaff et al. (2001) Acta Cryst. D57,

5 Output from substructure determination If substructure coordinates are found, usually all positions are determined accurately. Indicators of a correct solution: CCweak > 30% in SHELXD FOM > 1.0 in CRUNCH2 (both are conservative criteria for a correct solution)

6 Validating substructure detection A substructure is assumed to be solved if it is over a statistical threshold defined by the detection program (ie. CCweak > 30% or CRUNCH2 FOM > 1.0) Problem: Often, a substructure is correct, but the threshold is not reached. Solution: Run Bp3 in Check mode, to verify if a solution is complete/correct.

7 Speeding up substructure detection Run time CRUNCH mad_gere_3w mad_gere_3wn mad_gere_2w 1vk4-1 mad_gere_2wn sad_haptbr 2fg0-1 2pv4-1 2oh3-1 2evr-1 2ets-1 mad_cyanase_2w mad_cyanase_3w 2obp-1 sad_gere_peak sad_dna90 2nuj-1 2aml-1 sad_jia_peak sad_jia_infl 2oc5-1 mad_jia_2w mad_jia_3w 2o2x-1 mad_jia_4w siras_krel 1vjn-2 siras_rnase 2nlv-1 sad_caufd sad_dna360 sad_dna270 sad_toxdau 2o3l-1 sad_cyanase_peak siras_toxdau 2p4o-1 2avn-1 2o1q-1 2od5-1 2fea-1 siras_1hf8 sad_sav3 siras_gere siras_2b78 2ozj-1 2pr7-1 siras_2hba sad_dna180 sad_gere_infl 2etj-1 sad_cbm1 sad_crust 2gc9-1 2o62-1 siras_1e6i 2p97-1 2nwv-1 2ooj-1 2o08-1 2p7i-2 siras_2o0h 2pw4-1 2fna-1 2opk-1 sad_weiss360 2p97-2 2o8q-1 mad_pscp_3w mad_pscp_2w 1vjr-1 siras_3bpj sad_gilu 2fur-1 1z82-1 sad_pscp_peak 2o2z-1 siras_2b79 2ozg-1 2pim-1 2otm-1 siras_fcho2 sad_lyso2 sad_muts_peak siras_1e42 sad_weiss270 sad_weiss180 siras_2ahy 2fg0-2 2evr-2 2p7i-1 siras_2rkk Early termination Full run Plot of run-time vs data set for a full crunch2 run (green) and a crunch2 with early termination (red).

8 Important parameters in substructure detection The number of cycles run. The number of atoms to search for. Should be within 10-20% of actual number A first guess uses a probabilistic Matthew s coefficient The resolution cut-off: For MAD, look at signed anomalous difference correlation. For SAD, a first guess is high resolution limit.

9 BP3: Heavy atom refinement Can be used for SAD, MAD, S/MIR(AS). Refines atomic and error parameters. Outputs FOM, HL coefficients, PHIB to an MTZ file in original and inverted hand. Two modes of operation: normal and PHASe (fast phasing). Output from Bp3 should be input to a density modification program.

10 SAD functions in heavy atom refinement before BP3 Most heavy atom refinement programs use a Gaussian (or least squares) function in Bijvoet differences (ΔF = F + - F - ) (North, 1965), (Matthews, 1966). The calculated Bijvoet difference is determined based on a assumed value of F and α and the heavy atom structure factor model.

11 Deriving a likelihood function suitable for a SAD experiment Include effect of model and measurement errors and correlation between observed and calculated Bijvoet pairs. Required joint probability distribution is P( F + F, F - ; F + F - c+, F c )

12 Is my map good enough? Statistics from substructure phasing: Look at FOM from BP3. For SAD, look at Luzzati parameters. Refined occupancies. Statistics from density modification: Compare the contrast from hand and enantiomorph (output of solomon or shelxe). Does it look like a protein? (model visualization)

13 Improving the map Adjusting solvent content can improve the map after density modification. (Since the number of monomers is usually not known beforehand, neither is the solvent content.) If BP3 was run in fast mode, or SHELXE was run, a better map may result if BP3 is run in default mode. Use NCS averaging (see Crank/dm/Buccaneer demo on ccp4wiki.org).

14 Is my automatically built model correct? General comments for ARP/wARP, Buccaneer, and Resolve: What fraction of residues have been built? How long is the longest peptide built? What fraction of amino acids built have sequence docked?

15 Testing Crank s robustness Fraction model built vlm-1.log 2fcl-1.log 2prx-1.log siras_2o0h.log sad_weiss180.log sad_lyso2.log sad_muts_peak.log 2osd-1.log siras_3bpj.log sad_crust.log 2nwv-1.log siras_1e42.log 2ozj-1.log 2prx-2.log 2pg3-1.log 2pg4-1.log sad_gere_infl.log siras_gere.log 1vjn-2.log 2etd-1.log siras_2b79.log sad_gere_peak.log 2okc-1.log mad_gere_2w.log 1z82-1.log mad_gere_3w.log mad_gere_2wn.log 2okf-1.log 2fna-1.log 2ffj-1.log 2pv4-1.log 2o1q-1.log 2o7t-1.log 2o08-1.log 2o3l-1.log mad_gere_3wn.log 1vk4-1.log 2oh3-1.log 2avn-1.log 2nlv-1.log 2nuj-1.log 2p7i-2.log sad_insulin.log 2ozg-1.log mad_pscp_3w.log 2opk-1.log 2p4o-1.log 2p97-1.log mad_jia_4w.log 2ets-1.log sad_weiss270.log sad_sav3.log 2obp-1.log 2o2x-1.log 2fea-1.log 2o2z-1.log sad_weiss360.log 2o62-1.log 2aml-1.log sad_pscp_peak.log mad_cyanase_2w.log 2od5-1.log 1zy9-1.log mad_pscp_2w.log 2otm-1.log 2pw4-1.log 2oc5-1.log 1vjr-1.log sad_jia_infl.log sad_jia_peak.log mad_jia_2w.log 2gc9-1.log 2ooj-1.log sad_gilu.log sad_haptbr.log mad_jia_3w.log 2pr7-1.log 1zyb-1.log 2ppv-1.log mad_cyanase_3w.log sad_cbm1.log 2p97-2.log 2fur-1.log 2fg0-1.log 2evr-1.log siras_krel.log 2etv-1.log 0 Plot of fraction built versus data set by just varying one model building program (red) versus another (green). Pipeline 1 Pipeline 6

16 SAD function in model refinement Current functions in REFMAC: No prior phase information (Rice function) (Murshudov et al.,1997), (Bricogne and Irwin, 1996), (Pannu and Read, 1996) Prior phase information used indirectly in the form of Hendrickson-Lattman coefficients (MLHL) (Pannu et al., 1998)

17 Shortcomings of MLHL Dependent on where you obtained your Hendrickson-Lattman coefficients. Assumes that your prior phase information is independent from your model phases! Benefit: General approach for all experiments (MAD, SAD, MIRAS).

18 Tests of SAD function in refinement The SAD function was tested on many real SAD data sets (various anomalous signals and resolution ranges) against MLHL and RICE function in automated model building with iterative e refinement in ARP/wARP + REFMAC. Input to ARP/wARP an REFMAC created with CRANK using CRUNCH2 or SHELXD, BP3 and DM. Skubak et al. (2004,2005) 2005) Acta Cryst D.

19 Map correl. Results Map correl Resol Resol. Rice function MLHL function Map correl Green: % built Yellow: 50 80% built 0.5 Red: 20 50% built SAD function Resol. Black: 0-20% built

20 Cases in the ShelxC/D/E pipeline with differences * between SAD and RICE in warp * + Refmac Resol. Anom. Experiment Time Residues (Å) atoms (min) RICE/SAD/FINAL MutS Se SAD (peak) /1093/1600 subtilisin Ca, S SAD /259/275 thioesterase Se SAD (infl) /542/572 gere Se MAD(p/i) /110/444 cyanase Se MAD (p/i) /669/1560 thioesterase I Br SAD(peak) /431/462 * 10 warp cycles. # Total lcrank time for SAD target in Refmac.

21 Remarks on current Crank usage With a sufficient anomalous signal and resolution, structures can be solved automatically. When structures can not, first determine which step has failed (ie. Was a substructure found? Do I know the number of monomers? Etc.) Crank attempts to make re-running running steps easier.

22 Future developments Improved density modification using data directly. Using SIRAS data directly in model refinement and multivariate SIRAS phasing. Joint ligand refinement. Multivariate F A.

23 Developments in phase combination/density modification Traditional density modification attempts to combine a density modified map with experimental phases. Rather than assume independence, d we attempt to model the correlation between the modified map and the original phases directly. We have developed a phase combination algorithm to consider the modified structure factors, SAD data and SAD heavy atom substructures together in a multivariate fashion.

24 Preliminary results After phasing Current Solomon Multivariate ss 65.98/ / /0.735 subtilisin 65.67/ / /0.509 gere 70.98/ / /0.670 T4 bact / / /0.457 rnase 77.02/ / /0.312

25 FA estimation First step in solving a structure by SAD/MAD or SIRAS is to determine F A values. A F A is the structure factor amplitude corresponding to the substructure to input to direct methods programs (i.e. (ie SHELXD or CRUNCH2)

26 Current F A estimation F A is currently estimated by F + - F - for SAD data. Direct method programs are very sensitive to F A values. Improving estimates can improve hit rates of direct methods and solve things that can not previously been solved.

27 Multivariate SAD equation E( F A, F +, F - ) = F A P( F A, α A, F +, α +, F -, α - ) d F A dα A dα + dα - Giacovazzo previously proposed multivariate i t F A estimation, with an implementation assuming Bijvoet phases are equal. An equation can be obtained without the equal phase assumption requiring only one numerical integration. The equation has been implemented which reduces to Giacovazzo s equation if Bijvoet phases are equal.

28 Test cases: Correlations with final calculated E s Reso Anom f Corr Corr Atom ΔE E multi Ferrodoxin 0.94 Fe Thioesterase 2.5 Se Lyso S Lyso S DNA P DNA P

29 Joint refinement of structure, structure-ligand complexes Use multivariate function to refine together structure and structure-ligand complexes Motivation: refining (isomorphic) data sets together could highlight the difference between the structure and provide better maps of the ligand.

30 Test case of ligand cocktail in ARP/wARP (a) Correct ligand (CNA) (b) PUI (c) HJK (d) YUH (e) PUK (f) ZIP () (g) ZIK

31 Results Function Input Ligand 1 st choice 2 nd choice Current none PUI CNA Multivariate none CNA PUI Current CNA PUI CNA Multivariate CNA CNA PUI Current PUI PUI CNA Multivariate PUI CNA PUI Recall: CNA is correct ligand, PUI is incorrect

32 Multivariate SIRAS function for phasing and model refinement Currently in BP3 and SHARP, anomalous information is added for SIRAS and MAD by multiplying by a Gaussian term of Bijvoet differences (Thus, assuming independence with isomorphism term.) This isomorphic term also assumes uncorrelated errors. Better results may be obtained by deriving a multivariate function for SIRAS modelling the correlation amongst data sets.

33 Phasing function Refinement function Results Correctly built residues 2o0h 2b78 2b79 univariate Rice univariate i MLHL univariate Multi. SIRAS multivariate Rice multivariate MLHL multivariate Multi. SIRAS

34 Future developments MAD is NOT MIR a multivariate likelihood MAD function in phasing and model refinement. Multi-crystal support in Refmac for dealing with (severe) non-isomorphism in SIRAS and with radiation damage.

35 Availability Crank works under Linux, MAC OS, Windows and is free software. Crank is available in CCP4 version or from nl/software/crank/ Please use version 1.1 or higher! Crank k130( (to be released din CCP4612) 6.1.2) has annotated logfiles for displaying results.

36 Acknowledgements All dataset contributors (JCSG, Z. Dauter, M.Weiss, C.Mueller-Dieckmann) Garib Murshudov, Kevin Cowtan, George Sheldrick, Victor Lamzin, Tassos Perrakis, Gerrit Langer, Charles Ballard, Francois Remacle, Peter Briggs, Martyn Winn Cyttron