Volcano Plot parameters in JMP Pro

[#1]

We don't have JMP Clinical or Genomic so we're using JMP Pro to create a volcano plot according to this article:

How to Build a Volcano Plot in JMP! 

Most volcano plot differences refer to the x-axis as fold change or log2(fold change). The terminology in the article uses Difference. It also indicated to use Difference only as opposed to Relative Practical Difference.

In the wiki article on fold change, there is also a reference to relative change. 

In JMP context, how do all these terms relate to fold change or log2(fold change)? 

[#2]

Hi @mjmg ,

Welcome to the community!

Fold Change and Log2 (fold change) are both difference measures (so equivalent to differences in the blog post) that ranges from negative to positive with a zero center. So that would go on the x-axis.when creating an volcano plot.

Although Volcano Plots are a statistical plot typically used for genomic data analysis visualization, it could be more generally applied to other data and industries. We wanted to generalize its application and to start from from a JMP Platform, like Response Screening, to get those differences and log (p-values) for a large set of data in order to make a volcano plot.

If you do not have fold change or log2(fold change) then you can use the Response Screening Platform to generate them. I would start with log2 of the data before using response screening. Then when you chose compare means, they will be the log2(fold change) values.

Does that help?

Ultimately, it does not matter where you get.generate the fold changes or p-values from. You can make the plot in JMP using graph builder.

[#3]

@Chris_Kirchberg thanks for some of the efforts in making volcano plots more mainstream, I'm still a bit confused with the terminology. Hopefully your team can make the generation of volcano plots more streamlined or better explained.

Mainly I need to find the direction of the data generated by JMP volcano plot (whether it is ControlGroup-NotControlGroup or NotControlGroup-ControlGroup; is ControlGroup on the left or on the right side, etc...), set up fold change or differences thresholds in the x-axis that would be comparable to published references (x1 or x2 fold change or log2(fold change) = +/-1). The explanation is not that clear from the tutorial when the term Differences is used and is it equivalent to Fold Change or log2(Fold Change)?.

My data is not log2 transformed, but I'm working on normalized (against a reference), centered and autoscaled data. So it is better to have log2 transform on my data even if its normalized, centered and autoscaled already and include log2 transform in the preprocessing process so the Difference parameter is directly equivalent to log2 Fold Change?

[#4 · Accepted solution]

@mjmg 

The term Difference (or change) is equivalent to fold change or log2(fold change).

One could say that fold change is a type of difference (one value compared to another by using a ratio of one over the other) but the range of fold change is usually not suitable for a volcano plot (0 to infinity with the value of 1 representing no difference, so the scale of 0 to 1 is compressed compared to 1 to a very large positive number). Usually people use log2 (or log10) because it puts the range into - infinity to infinity with 0 being the center point on the graph or representing no difference or change between two treatments or control vs. treatment.

I think that in your situation, if you already have centered, scaled and normalized data, it should suffice to being put on the x-axis. The issue will be a matter if the left size is compressed compared to the right side of the plot. More of an aesthetic issue than an analytical one. The main concept is that there is a 0 or center point in the x-axis that represents no difference/change between control and treatment (or a fold change of 1). This is why people usually take the log of the data. Easier to filter and easier to plot.

As for which side control group should be, that is up to you. Your choice.  Most people have control listed like this:  treatment/control or the log2 treatment - log 2 control.  That way data on the right of the plot are greater in treatment and data on the left are greater in control.

[#5]

Thanks for the good explanation. I've seen some papers use only the threshold on the y-axis or significant figures. For JMP with Difference terminology, what would be good thresholds on the x-axis?

[#6]

@mjmg Some people call those threshold lines in the x-axis goal posts but I think that is a misnomer. You want what is outside those posts (right or left), What they should be is up to the analysis or data.  In gene expression, people use 1.5 to 2 fold change. In reality, it is what ever is practical or has meaning. I know from experience that some genes need only a 1.2 fold change in expression to have an impact on what the cells are doing (usually changes in metabolic pathways or environmental signaling pathways).

The same is true for the y-axis threshold. A lot of literature uses a p-value cutoff of 0.05, but there is no real meaning behind that cutoff. It is arbitrary.  I choose depending on the risk and/or need of the experiment. If I am screening genes, I probably want a more relaxed threshold on the y-axis (some say 0.1 is good enough) and then look at 1.5 fold change on the x-axis. That way I am less likely to miss something that could impact the system I am studying.  I then look at networks and pathways that are impacted by that large set to help interpret and then narrow down the list of genes I want to do a follow up experiment to better understand and confirm my observations. Maybe expand that experiment to test more factors that could influence my system.