interpretation of experimental scores wgEncodeRegTfbsClusteredV2

[Robert O'Connor]

Hi –

 

I am wanting to determine if the TFBS experimental scores in wgEncodeRegTfbsClusteredV2 file are useful in some way for filtering the TFBS data or if they are merely useful as a browser attribute.

I understand that the score field indicates how dark the peak will be displayed in the browser (0-1000), with 7 being lowest non-zero score.

I also understand that the individual experimental scores (expScores) mapping to experimental ids (expIds) are proportional to the signal strength measured in the Chip-seq experiment.

 

What could one say about the following two observations from the file for experiment Id ‘3’?

 

chrom   chromStart         chromEnd           name    score     strand                   …             expIds                  expScores

chr16     9813018                9813202                NRSF     7              .                               …             0,1,2,3,4,…          0,0,0,7,0              

chr16     9930044                9930318                NRSF     1000       .                               …             0,1,2,3,4,…          353,1000,806,248,0

 

Is it that both are equally valid but the ‘248’ experimental score site has a greater strength or frequency of association with the genome region than the ‘7’ experimental score site for experiment Id ‘3’?

Thank you for the help.

 

Bob

 

 

[Kate Rosenbloom]

Hello Bob,

Yes, the scores in the TFBS Clusters track do indeed reflect signal
strength so you may find filtering on score of some use. And your
observations based on examples below are correct. Youcan confirm by
displaying the cluster details in the browser -- navigate to the regions
of interest and click through to see the details of cluster there, e.g.:

http://genome.ucsc.edu/cgi-bin/hgc?db=hg19&c=chr16&o=9813018&t=9813202&g=wgEncodeRegTfbsClusteredV2&i=NRSF

In this cluster only a single cell type (HepG2, cell type #4) showed
enrichment, and weakly (signal strength of 7)

http://genome.ucsc.edu/cgi-bin/hgc?db=hg19&c=chr16&o=9930044&t=9930318&g=wgEncodeRegTfbsClusteredV2&i=NRSF

While in this cluster, all 9 cell types assayed for NRSF showed
enrichment, with HepG2 signal at 248.

The individual experiment scores (expScores) are derived from the signal
strength in the underlying datasets, normalized to lie within the 0-1000
range. And the cluster score is simply the maximum expScore contributing
to the cluster.

Hope this helps,

Kate
----
Kate Rosenbloom
UCSC Genome Bioinformatics


On 3/21/13 4:06 PM, Robert O'Connor wrote:
>
> Hi �

>
> I am wanting to determine if the TFBS experimental scores in
> wgEncodeRegTfbsClusteredV2 file are useful in some way for filtering
> the TFBS data or if they are merely useful as a browser attribute.
>
> I understand that the score field indicates how dark the peak will be
> displayed in the browser (0-1000), with 7 being lowest non-zero score.
>
> I also understand that the individual experimental scores (expScores)
> mapping to experimental ids (expIds) are proportional to the signal
> strength measured in the Chip-seq experiment.
>
> What could one say about the following two observations from the file

> for experiment Id �3�?
>
> chrom chromStart chromEnd name score strand � expIds expScores
>
> chr16 9813018 9813202 NRSF 7 . � 0,1,2,3,4,� 0,0,0,7,0
>
> chr16 9930044 9930318 NRSF 1000 . � 0,1,2,3,4,� 353,1000,806,248,0
>
> Is it that both are equally valid but the �248� experimental score

> site has a greater strength or frequency of association with the

> genome region than the �7� experimental score site for experiment Id �3�?

>
> Thank you for the help.
>
> Bob
>

> --
>
>
>

[Robert O'Connor]

Hi Kate,

Would it be possible to get an explanation/documentation about how the normalization from original signal strength value (signalValue) to the normalized value is accomplished?
I find myself needing to explain it to others.
Also it there any consensus on how a 'weak' interaction would be defined with respect to the score?
Thanks for all the help,

Bob

-----Original Message-----
From: Kate Rosenbloom [mailto:ka...@soe.ucsc.edu]
Sent: Wednesday, March 27, 2013 5:07 PM
To: Robert O'Connor
Cc: gen...@soe.ucsc.edu
Subject: Re: [genome] interpretation of experimental scores wgEncodeRegTfbsClusteredV2

Hello Bob,

Yes, the scores in the TFBS Clusters track do indeed reflect signal strength so you may find filtering on score of some use. And your observations based on examples below are correct. Youcan confirm by displaying the cluster details in the browser -- navigate to the regions of interest and click through to see the details of cluster there, e.g.:

http://genome.ucsc.edu/cgi-bin/hgc?db=hg19&c=chr16&o=9813018&t=9813202&g=wgEncodeRegTfbsClusteredV2&i=NRSF

In this cluster only a single cell type (HepG2, cell type #4) showed enrichment, and weakly (signal strength of 7)

http://genome.ucsc.edu/cgi-bin/hgc?db=hg19&c=chr16&o=9930044&t=9930318&g=wgEncodeRegTfbsClusteredV2&i=NRSF

While in this cluster, all 9 cell types assayed for NRSF showed enrichment, with HepG2 signal at 248.

The individual experiment scores (expScores) are derived from the signal strength in the underlying datasets, normalized to lie within the 0-1000 range. And the cluster score is simply the maximum expScore contributing to the cluster.

Hope this helps,

Kate
----
Kate Rosenbloom
UCSC Genome Bioinformatics


On 3/21/13 4:06 PM, Robert O'Connor wrote:
>

> Hi -

>
> I am wanting to determine if the TFBS experimental scores in
> wgEncodeRegTfbsClusteredV2 file are useful in some way for filtering
> the TFBS data or if they are merely useful as a browser attribute.
>
> I understand that the score field indicates how dark the peak will be
> displayed in the browser (0-1000), with 7 being lowest non-zero score.
>
> I also understand that the individual experimental scores (expScores)
> mapping to experimental ids (expIds) are proportional to the signal
> strength measured in the Chip-seq experiment.
>
> What could one say about the following two observations from the file

> for experiment Id '3'?
>
> chrom chromStart chromEnd name score strand ... expIds expScores
>
> chr16 9813018 9813202 NRSF 7 . ... 0,1,2,3,4,... 0,0,0,7,0
>
> chr16 9930044 9930318 NRSF 1000 . ... 0,1,2,3,4,... 353,1000,806,248,0
>
> Is it that both are equally valid but the '248' experimental score

> site has a greater strength or frequency of association with the

> genome region than the '7' experimental score site for experiment Id '3'?

[Brooke Rhead]

Hi Bob,

Input signal values are multiplied by a normalization factor calculated
as the ratio of the maximum score value (1000) to the signal value at 1
standard deviation from the mean, with values exceeding 1000 capped at
1000. This has the effect of distributing scores up to mean + 1std
across the score range, but assigning all above to the max score.

We are not aware of a standard threshold for weak interactions.

If you have further questions, please feel free to contact us again at
gen...@soe.ucsc.edu.

--
Brooke Rhead
UCSC Genome Bioinformatics Group