Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination
description
Transcript of Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination
![Page 1: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/1.jpg)
Ameet Soni* and Jude ShavlikDept. of Computer SciencesDept. of Biostatistics and Medical InformaticsCraig BingmanDept. of BiochemistryCenter for Eukaryotic Structural Genomics
Presented at the ACM International Conference on Bioinformatics and Computational Biology 2010
Guiding Belief Propagationusing Domain Knowledge for
Protein-Structure Determination
![Page 2: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/2.jpg)
2
Protein Structure Determination
Proteins essential to most cellular function Structural support Catalysis/enzymatic activity Cell signaling
Protein structures determine function
X-ray crystallography is main technique for determining structures
2
![Page 3: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/3.jpg)
3
X-ray Crystallography: Background
Electron-DensityMap (3D Image)
Interpret
Protein Crystal
X-ray Beam
Protein Structure
3
Diffraction pattern
FFT
Collect
![Page 4: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/4.jpg)
4
Task Overview Given:
A protein sequence Electron-density map
(EDM) of protein
Do: Automatically produce a
protein structure (or trace) that is All atom Physically feasible
4
SAVRVGLAIM...
![Page 5: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/5.jpg)
5
Challenges & Related Work
1 Å 2 Å 3 Å 4 Å
Our Method: ACMI
5
ARP/wARPTEXTAL & RESOLVE
![Page 6: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/6.jpg)
6 ACMI Overview6
- Background
- Inference in ACMI-BP- Guiding Belief Propagation- Experiments & Results
![Page 7: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/7.jpg)
7
Our Technique: ACMIPerform Local Match Apply Global
Constraints Sample Structure
ACMI-SH ACMI-BP ACMI-PF
7
pk+1(b )k+1*1pk+1(b )k+1*2
pk+1(b )k+1*M
…
bk
bk-1
bk+1*1…M
a priori probability of
each AA’s location
marginal probabilityof each AA’s location
all-atom protein structures
![Page 8: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/8.jpg)
8
Previous Work [DiMaio et al, 2007]8
![Page 9: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/9.jpg)
9
ACMI FrameworkPerform Local Match Apply Global
Constraints Sample Structure
ACMI-SH ACMI-BP ACMI-PF
9
pk+1(b )k+1*1pk+1(b )k+1*2
pk+1(b )k+1*M
…
bk
bk-1
bk+1*1…M
a priori probability of
each AA’s location
marginal probabilityof each AA’s locationall-atom protein structures
![Page 10: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/10.jpg)
10
Inference in ACMI-BP10
- Background- ACMI Overview
- Guiding Belief Propagation- Experiments & Results
![Page 11: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/11.jpg)
11
ACMI-BP11
ACMI models the probability of all possible traces using a pairwise Markov Random Field (MRF)
LEU SERGLY LYSALA
![Page 12: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/12.jpg)
12
ACMI-BP: Pairwise Markov Field
LEU SERGLY LYSALA
Model ties adjacency constraints, occupancy constraints, and Phase 1 priors
12
![Page 13: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/13.jpg)
13
Approximate Inference P(U|M) intractable to calculate, maximize
exactly
ACMI-BP uses Loopy Belief Propagation (BP) Local, message-passing scheme Distributes evidence between nodes Approximates marginal probabilities if graph
has cycles
13
![Page 14: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/14.jpg)
14
ACMI-BP: Loopy Belief Propagation
LYS31 LEU32
mLYS31→LEU32
pLEU32pLYS31
14
![Page 15: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/15.jpg)
15
ACMI-BP: Loopy Belief Propagation
LYS31 LEU32
mLEU32→LEU31
pLEU32pLYS31
15
![Page 16: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/16.jpg)
16
Guiding Belief Propagation16
- Background- ACMI Overview- Inference in ACMI-BP
- Experiments & Results
![Page 17: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/17.jpg)
Best case: wasted resources Worst case: poor information given more influence
Message Scheduling17
SERLYSALA
Key design choice: message-passing schedule When BP is approximate, ordering affects
solution[Elidan et al, 2006]
ACMI-BP uses a naïve, round-robin schedule
![Page 18: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/18.jpg)
18
Using Domain Knowledge18
Idea: use expert to assign importance of messages
Biochemist insight: well-structured regions of protein correlate with strong features in density map eg, helices/strands have stable conformations
Protein disorder - regions of a structure that are unstable/hard to define ACMI-BP can use disorder to decide
importance Accurate predictors exist based on sequence
alone
![Page 19: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/19.jpg)
19
Guided ACMI-BP19
![Page 20: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/20.jpg)
20
Related Work Assumption: messages with largest
change in value are more useful
Residual Belief Propagation [Elidan et al, UAI 2006] Calculates residual factor for each node
Each iteration, highest residual node passes messages
General BP technique
20
![Page 21: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/21.jpg)
21
Experiments & Results21
- Background- ACMI Overview- Inference in ACMI-BP- Guiding Belief Propagation
![Page 22: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/22.jpg)
22
Message Schedulers Tested22
Our previous technique: naive, round robin (BP)
Our proposed technique: Guidance using disorder prediction (DOBP) Disorder prediction using DisEMBL [Linding et al,
2003] Prioritize residues with high stability (ie, low
disorder)
Residual factor (RBP) [Elidan et al, 2006]
![Page 23: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/23.jpg)
23
Experimental Methodology Run whole ACMI pipeline
Phase 1: Local amino-acid finder (prior probabilities)
Phase 2: Either BP, DOBP, or RBP Phase 3: Sample all-atom structures from
Phase 2 results
Test set of 10 poor-resolution electron-density maps From UW Center for Eukaryotic Structural
Genomics Deemed the most difficult of a large set of
proteins
23
![Page 24: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/24.jpg)
24
ACMI-BP Marginal Accuracy24
![Page 25: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/25.jpg)
25
ACMI-BP Marginal Accuracy25
![Page 26: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/26.jpg)
27
Protein Structure Results27
Do these better marginals produce more accurate protein structures?
RBP fails to produce structures in ACMI-PF Marginals are high in entropy (28.48 vs 5.31) Insufficient sampling of correct locations
![Page 27: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/27.jpg)
28
Conclusions Our contribution: framework for utilizing
domain knowledge in BP message scheduling General technique for belief propagation Alternative to information-based techniques
Our technique improves inference in ACMI Disorder prediction used in our framework Residual-based technique fails
Future directions
28
![Page 28: Guiding Belief Propagation using Domain Knowledge for Protein-Structure Determination](https://reader035.fdocuments.us/reader035/viewer/2022062810/56815d7e550346895dcb8bbf/html5/thumbnails/28.jpg)
29
Phillips Laboratory at UW - Madison UW Center for Eukaryotic Structural
Genomics (CESG)
NLM R01-LM008796 NLM Training Grant T15-LM007359 NIH Protein Structure Initiative Grant
GM074901
Thank you!
Acknowledgements29