Incorporating biological information into network inference using structured shrinkage

Duration: 20 mins 51 secs

Share this media item:

Embed this media item:

<iframe width="_width_" height="_height_" src="https://sms.cam.ac.uk/media/2310566/embed" frameborder="0" scrolling="no" allowfullscreen></iframe>

Choose size:

About this item

Available Formats

About this item

Description:	Leday, G (MRC Biostatistics Unit) Friday 26th August 2016 -10:40 to 11:00


Created:	2016-08-31 16:58
Collection:	Theoretical Foundations for Statistical Network Analysis
Publisher:	Isaac Newton Institute
Copyright:	Leday, G
Language:	eng (English)


Abstract:	High-throughput biotechnologies such as microarrays provide the opportunity to study theinterplay between molecular entities, which is central to the understanding of disease biology.The statistical description and analysis of this interplay is naturally carried out with Gaussiangraphical models in which nodes represent molecular variables and edges between them representinteractions. Inferring the edge set is, however, a challenging task as the number of parametersto estimate easily is much larger than the sample size. A conventional remedy is to regularize orpenalize the model likelihood. In network models, this is often done locally in the neighbourhoodof each node. However, estimation of the many regularization parameters is often dicult andcan result in large statistical uncertainties. We show how to combine local regularization withglobal shrinkage of the regularization parameters, via empirical Bayes (EB), to borrow strengthbetween nodes and improve inference. Furthermore, we show how one can use EB so the level ofregularization may dier across an arbitrary number of predened groups of interactions. Suchauxiliary information is often available in Biology. It is shown that accurate prior information cangreatly improve the reconstruction of the network, but need not harm the reconstruction if wrong.

Abstract:

High-throughput biotechnologies such as microarrays provide the opportunity to study theinterplay between molecular entities, which is central to the understanding of disease biology.The statistical description and analysis of this interplay is naturally carried out with Gaussiangraphical models in which nodes represent molecular variables and edges between them representinteractions. Inferring the edge set is, however, a challenging task as the number of parametersto estimate easily is much larger than the sample size. A conventional remedy is to regularize orpenalize the model likelihood. In network models, this is often done locally in the neighbourhoodof each node. However, estimation of the many regularization parameters is often dicult andcan result in large statistical uncertainties. We show how to combine local regularization withglobal shrinkage of the regularization parameters, via empirical Bayes (EB), to borrow strengthbetween nodes and improve inference. Furthermore, we show how one can use EB so the level ofregularization may dier across an arbitrary number of predened groups of interactions. Suchauxiliary information is often available in Biology. It is shown that accurate prior information cangreatly improve the reconstruction of the network, but need not harm the reconstruction if wrong.

Available Formats

Format	Quality	Bitrate	Size
MPEG-4 Video	640x360	1.94 Mbits/sec	303.59 MB	View	Download
WebM	640x360	980.62 kbits/sec	149.87 MB	View	Download
iPod Video	480x270	522.51 kbits/sec	79.79 MB	View	Download
MP3	44100 Hz	249.85 kbits/sec	38.19 MB	Listen	Download
Auto *	(Allows browser to choose a format it supports)