Top Banner
network deconvolution as a general method to distinguish direct dependencies in networks MIT group; Accepted Jun. 2013; Nature Biotechnology Presented by Haicang Zhang Feb.24 2013
14

network deconvolution as a general method to distinguish direct dependencies in networks

Feb 25, 2016

Download

Documents

sahara

network deconvolution as a general method to distinguish direct dependencies in networks . MIT group; Accepted Jun. 2013; Nature Biotechnology Presented by Haicang Zhang Feb.24 2013. Outline. Motivation Basic idea of Method Applications and results gene regulatory networks - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: network  deconvolution  as a general method to distinguish direct dependencies in networks

network deconvolution as a general method to distinguish direct dependencies in networks

MIT group; Accepted Jun. 2013; Nature Biotechnology

Presented by Haicang ZhangFeb.24 2013

Page 2: network  deconvolution  as a general method to distinguish direct dependencies in networks

Outline• Motivation• Basic idea of Method• Applications and results

– gene regulatory networks– protein structural constrains – co-authorship collaboration relationships

• Methods in detail – framework – convergence – computational complexity

• Discussion

Page 3: network  deconvolution  as a general method to distinguish direct dependencies in networks

Motivation

• Networks are usually to represent the interdependencies among variables.

• Indirect dependencies occurs owing to transitive effects of correlations.

• It is necessary to separate the direct dependence from the indirect ones.

Page 4: network  deconvolution  as a general method to distinguish direct dependencies in networks

Motivation

• Current method– partial correlations method– graphical models– other methods

Page 5: network  deconvolution  as a general method to distinguish direct dependencies in networks

Basic idea of method

• i

Page 6: network  deconvolution  as a general method to distinguish direct dependencies in networks

Basic idea of method

• i

Page 7: network  deconvolution  as a general method to distinguish direct dependencies in networks

Applications and results- protein structural constraints

• i

Page 8: network  deconvolution  as a general method to distinguish direct dependencies in networks

Applications and results- protein structural constraints

• i

Page 9: network  deconvolution  as a general method to distinguish direct dependencies in networks

Methods in detail-framework

• s

Page 10: network  deconvolution  as a general method to distinguish direct dependencies in networks

Methods in detail-intuition

• The intuition of this method– Network de-convolution can be viewed as a

nonlinear filter that is applied to eigenvalues of the observed dependency matrix.

– In general, ND decreases magnitudes of large positive eigenvalues of the observed matrix since transitivity inflate these positive eigenvalues.

Page 11: network  deconvolution  as a general method to distinguish direct dependencies in networks

Methods in detail-intuition

• i

Page 12: network  deconvolution  as a general method to distinguish direct dependencies in networks

Methods in detail-convergences

• converges iff the largest absolute value of eigenvalues of G_dir < 1.

• scale G_obs

• if beta < 1 , the largest absolute value of eigenvalues < 1

• Then,

Page 13: network  deconvolution  as a general method to distinguish direct dependencies in networks

Methods in detail-convergences

• i

Page 14: network  deconvolution  as a general method to distinguish direct dependencies in networks

Discussion

• Network de-convolution provides a general framework for computing direct dependencies in a network by use of observed similarities.

• It can recognize and remove spurious transitive edges due to indirect effects, decrease edge weights that are overestimated owing to indirect relationships, and assign edge weights corresponding to direct dependencies to the remaining edges