1. Forces and Prediction of Protein Structure Ming-Jing Hwang ( 黃明經 ) Institute of Biomedical Sciences Academia Sinica http://gln.ibms.sinica.edu.tw/

2. Science 2005

3. Sequence - Structure - Function MADWVTGKVTKVQ NWTDALFSLTVHAP VLPFTAGQFTKLGLE IDGERVQRAYSYVN SPDNPDLEFYLVTVP DGKLSPRLAALKPG DEVQVVSEAAGFFV LDEVPHCETLWMLA TGTAIGPYLSILR

4. Sequence/Structure Gap Current (May 15, 2007) entries in protein sequence and structure database:  SWISS-PROT/TREMBL : 267,354/4,361,897  PDB : 43,459 Sequence Structure

5. Structural Bioinformatics: Sequence/Structure Relationship All possible sequences of amino acids Protein sequences observed in nature Protein structures observed in nature 100 90 80 70 60 50 40 30 20 10 0 Percent Identity Twilight zone Midnight zone

6. Structure Prediction Methods 0 10 20 30 40 50 60 70 80 90 100 ab initio Fold recognition % sequence identity Homology modeling

7. Levinthal’s paradox (1969) If we assume three possible states for every flexible dihedral angle in the backbone of a 100-residue protein, the number of possible backbone configurations is 3 200. Even an incredibly fast computational or physical sampling in 10 -15 s would mean that a complete sampling would take 10 80 s, which exceeds the age of the universe by more than 60 orders of magnitude. Yet proteins fold in seconds or less! Berendsen

8. Energy landscapes of protein folding Borman, C&E News, 1998

9. Levitt ’ s lecture for S*S*

10. Levitt

12. Other factors Formation of 2nd elements Packing of 2nd elements Topologies of fold Metal/co-factor binding Disulfide bond …

13. Ab initio/new fold prediction Physics-based (laws of physics) Knowledge-based (rules of evolution)

14. Levitt

27. Molecular Mechanics (Force Field)

28. Levitt

30. 1-microsecond MD simulation 980ns - villin headpiece - 36 a.a. - 3000 H2O - 12,000 atoms - 256 CPUs (CRAY) -~4 months - single trajectory Duan & Kollman, 1998

31. Protein folding by MD PROTEIN FOLDING: A Glimpse of the Holy Grail? Herman J. C. Berendsen * * "The Grail had many different manifestations throughout its long history, and many have claimed to possess it or its like". We might have seen a glimpse of it, but the brave knights must prepare for a long pursuit.

32. Massively distributed computing SETI@home: SETI@home Folding@home Distributed folding Sengent’s drug design FightAIDS@home …

33. Letters to nature (2002) - engineered protein (BBA5) - zinc finger fold (w/o metal) - 23 a.a. - solvation model - thousands of trajectories each of 5-20 ns, totaling 700  s - Folding@home - 30,000 internet volunteers - several months, or ~a million CPU days of simulation Massively distributed computing

34. Energy landscapes of protein folding Borman, C&E News, 1998

35. Protein-folding prediction technique CGU: Convex Global Underestimation - K. Dill ’ s group

36. Challenges of physics-based methods Simulation time scale Computing power Sampling Accuracy of energy functions

37. Structure Prediction Methods 0 10 20 30 40 50 60 70 80 90 100 ab initio Fold recognition % sequence identity Homology modeling

38. Flowchart of homology (comparative) modeling From Marti-Renom et al.Marti-Renom et al.

39. Fold recognition Find, from a library of folds, the 3D template that accommodates the target sequence best. Also known as “ threading ” or “ inverse folding ” Useful for twilight-zone sequences

40. Fold recognition (aligning sequence to structure) (David Shortle, 2000)

41. 3D->1D score

42. On X-ray, NMR, and computed models

43. (Rost, 1996)

44. Marti-Renom et al. (2000) Reliability and uses of comparative models

45. Pitfalls of comparative modeling Cannot correct alignment errors More similar to template than to true structure Cannot predict novel folds

46. Ab initio/new fold prediction Physics-based (laws of physics) Knowledge-based (rules of evolution)

47. From 1D  2D  3D SISAY VQGTEACRHLTNLVNH LGINCRGSSQCGLSGGNLMVRIRDQACGNQGQTWCPGERRAKVCGTGNSISAY VQSTNNCISGTEACRHLTNLVNHGCRVCGSDPLYAGNDVSRGQLTVNYVNSC Tertiary Primary Secondary (fragment) fragment assembly seq. to str. mapping

48. CASP Experiments

49. One lab dominated in CASP4 One group dominates the ab initio (knowledge-based) prediction

50. Some CASP4 successes Baker ’ s group

51. Ab initio structure prediction server

52. The prediction of protein structure from amino acid sequence is a grand challenge of computational molecular biology. By using a combination of improved low- and high- resolution conformational sampling methods, improved atomically detailed potential functions that capture the jigsaw puzzle–like packing of protein cores, and high- performance computing, high-resolution structure prediction (<1.5 angstroms) can be achieved for small protein domains (<85 residues). The primary bottleneck to consistent high-resolution prediction appears to be conformational sampling. Toward High-Resolution de Novo Structure Prediction for Small Proteins --Philip Bradley, Kira M. S. Misura, David Baker (Science 2005)

53. Science 2003 3D to 1D?

54. A computer-designed protein (93 aa) with 1.2 A resolution

55. Structure prediction servers http://bioinfo.pl/cafasp/list.html

56. Hybrid approach for solving macromolecular complex structures

57. Thank You!