The Quixote Project: a pioneering work in managing Computational Chemistry research data

Transcription

1 1 The Quixote Project: a pioneering work in managing Computational Chemistry research data Pablo Echenique echenique.p@gmail.com

2 2 The protein folding problem Folding Native state Human Albumin sequence 585 amino acid residues, 4496 atoms

3 2 The protein folding problem Folding Native state human protein sequences Human Albumin sequence 585 amino acid residues, 4496 atoms 1385 human protein structures * Data extracted on 11/14/2008

4 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, 2009.

5 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms

6 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms During: ~1ms = 10-3s

7 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms During: ~1ms = 10-3s Using Anton, a specific-purpose computer, and a code specially adapted to it

8 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms During: ~1ms = 10-3s Using Anton, a specific-purpose computer, and a code specially adapted to it They report calculations in larger systems of a size similar to the typical one and achieving 5μs/day

9 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms During: ~1ms = 10-3s Using Anton, a specific-purpose computer, and a code specially adapted to it They report calculations in larger systems of a size similar to the typical one and achieving 5μs/day They see hundreds of folding events

10 The computational requirements of the protein folding problem 3 State-of-the-art example in a dedicated computer: D. E. Shaw et al., Millisecond-scale molecular dynamics simulations on Anton, Proceedings of the ACM/IEEE Conference on Supercomputing (SC09), November 14-20, residue protein with explicit water molecules: ~ atoms During: ~1ms = 10-3s Using Anton, a specific-purpose computer, and a code specially adapted to it They report calculations in larger systems of a size similar to the typical one and achieving 5μs/day They see hundreds of folding events Simulating a 106-atoms system for 1s in Anton would still take more than 500 wall-clock years

11 4 Present-day energy functions have to be improved Force fields comparison: S. Piana, K. Lindorff-Larsen, and D. E. Shaw, How Robust Are Protein Folding Simulations with Respect to Force Field Parameterization?, Biophys. J. 100 (2011)

12 4 Present-day energy functions have to be improved Force fields comparison: S. Piana, K. Lindorff-Larsen, and D. E. Shaw, How Robust Are Protein Folding Simulations with Respect to Force Field Parameterization?, Biophys. J. 100 (2011) residue protein with explicit water molecules: ~5 103 atoms

13 4 Present-day energy functions have to be improved Force fields comparison: S. Piana, K. Lindorff-Larsen, and D. E. Shaw, How Robust Are Protein Folding Simulations with Respect to Force Field Parameterization?, Biophys. J. 100 (2011) residue protein with explicit water molecules: ~5 103 atoms During: ~0.5ms (different runs for different force fields)

14 4 Present-day energy functions have to be improved Force fields comparison: S. Piana, K. Lindorff-Larsen, and D. E. Shaw, How Robust Are Protein Folding Simulations with Respect to Force Field Parameterization?, Biophys. J. 100 (2011) residue protein with explicit water molecules: ~5 103 atoms During: ~0.5ms (different runs for different force fields) Four force fields: 2 CHARMM variants and 2 AMBER ones

15 4 Present-day energy functions have to be improved Force fields comparison: S. Piana, K. Lindorff-Larsen, and D. E. Shaw, How Robust Are Protein Folding Simulations with Respect to Force Field Parameterization?, Biophys. J. 100 (2011) residue protein with explicit water molecules: ~5 103 atoms During: ~0.5ms (different runs for different force fields) Four force fields: 2 CHARMM variants and 2 AMBER ones Native state and folding rates are coincident between force fields, but folding pathways are not:

16 5 Present-day energy functions have to be improved Mecánica cuántica no relativista = química cuántica

17 5 Present-day energy functions have to be improved Mecánica cuántica no relativista = química cuántica Peptides

18 Where do you start? 6

19 Mmm, soon no human will be able to keep the pace 7

20 OK, I still want to do my study. How do I find and get the data? 8 To find and select the data

21 OK, I still want to do my study. How do I find and get the data? To find and select the data 8 To get the data

22 OK, I still want to do my study. How do I find and get the data? To find and select the data 8 To get the data

23 9 2011? Internet??!?

24 9 2011? Internet??!??!?

25 And... what if I am interested in crystal structures of proteins? 10

28 So... why not in computational chemistry? 13 Why not in computational chemistry?

29 13 So... why not in computational chemistry? Why not in computational chemistry? Because there is little data and it can be searched in a 20th Century fashion? Because nobody needs to do such a search?

30 13 So... why not in computational chemistry? Why not in computational chemistry? Because there is little data and it can be searched in a 20th Century fashion? Because nobody needs to do such a search? Because the data is cheaper to obtain? The data I am going to show later can be estimated to cost approximately 5000 (then), not including electricity or salaries.

31 13 So... why not in computational chemistry? Why not in computational chemistry? Because there is little data and it can be searched in a 20th Century fashion? Because nobody needs to do such a search? Because the data is cheaper to obtain? The data I am going to show later can be estimated to cost approximately 5000 (then), not including electricity or salaries. Because it is theory and not experiment? Quantum chemistry calculations have been shown to be more precise than experiments in several cases.

32 13 So... why not in computational chemistry? Why not in computational chemistry? Because there is little data and it can be searched in a 20th Century fashion? Because nobody needs to do such a search? Because the data is cheaper to obtain? The data I am going to show later can be estimated to cost approximately 5000 (then), not including electricity or salaries. Because it is theory and not experiment? Quantum chemistry calculations have been shown to be more precise than experiments in several cases. Because nobody has decided to build the infrastructure?

33 Quixote 14

34 15 Real use case P. Echenique and J. L. Alonso, Efficient model chemistries for peptides. I. General framework and a study of the heterolevel approximation in RHF and MP2 with Pople split-valence basis sets, Journal of Computational Chemistry 29 (2008)

35 Real use case 16

36 Real use case 17

37 Put the data in an Apache server and link from the paper 18

38 Put the data in an Apache server and link from the paper 19 Pros Instead of the CD, I can send the link to you.

39 Put the data in an Apache server and link from the paper 19 Pros Instead of the CD, I can send the link to you. Cons I have to maintain and curate the server. Persistence of the link? No metadata. No searchability and indexing (metadata-wise). No searchability, indexing or analysis (QC-wise). Little interoperability with other services.

40 Put the data in an Institutional Repository 20

41 Put the data in an Institutional Repository 21

42 22 Put the data in an Institutional Repository Pros Instead of the CD, I can send the link to you. The server is maintained and curated for me. The funding for the infrastructure is typically long-term. The link is persistent. Rich metadata. Searchability and indexing (metadata-wise). Interoperability with other services (metadata-wise).

43 22 Put the data in an Institutional Repository Pros Instead of the CD, I can send the link to you. The server is maintained and curated for me. The funding for the infrastructure is typically long-term. The link is persistent. Rich metadata. Searchability and indexing (metadata-wise). Interoperability with other services (metadata-wise). Cons No searchability, indexing or analysis (QC-wise). No interoperability with other services (QC-wise). Lack of specificity. Motivation to upload?

44 Put the data in a thematic repository 23

45 24 Put the data in an thematic repository Pros Instead of the CD, I can send the link to you. Rich metadata. Searchability and indexing (metadata-wise). Interoperability with other services (metadata-wise). Searchability, indexing or analysis (QC-wise). Interoperability with other services (QC-wise). Specific (and hence motivation?). The server is maintained and curated for me (hopefully). The link is persistent (hopefully).

46 24 Put the data in an thematic repository Pros Instead of the CD, I can send the link to you. Rich metadata. Searchability and indexing (metadata-wise). Interoperability with other services (metadata-wise). Searchability, indexing or analysis (QC-wise). Interoperability with other services (QC-wise). Specific (and hence motivation?). The server is maintained and curated for me (hopefully). The link is persistent (hopefully). Cons The funding for the infrastructure can be unstable.

47 25 Quixote versatile, lightweight, modular, open source,...

48 Chempound 26

49 Chempound 27

50 50 Some of the people involved José Luis Alonso UZ-BIFI Pablo de Castro SONEX Pablo Echenique CSIC-UZ-BIFI Jorge Peter Estrada Murray-Rust CSIC-UZ-ZCAM Cambridge Thank you for your attention