What is the Variance means in PCA plot ?

Question

[sabaghianamir70]

What is Those variance means exactly in this PCA plot

I had 2 paired end data, 1 normal and 1 treat and because i just had these two, i had some error for analyzing them and the results were wrong. So i turned them into single ends, and i got acceptable results. but i worried about this plot because of that 2%. can some one explain about that ?

Am i going to have problem with that 2% variance or it is ok ?

[andres.firrincieli]

I had 2 paired end data, 1 normal and 1 treat and because i just had these two

I do not really understand how your dataset is structured. If you are attempting a DEG analysis with just 1 replicate per condition, do not even bother doing that.

Am i going to have problem with that 2% variance or it is ok ?

However, let say you have 2 biological replicates per condition, I would be worried about that 98% variance; the differences across your conditions are so huge that you will not get any meaningful results from a DEG analysis (too many DEG).

[Kevin Blighe]

Indeed, that PCA bi-plot looks strange... Can you please show or explain your data processing steps, and outline your experiment, generally?

[sabaghianamir70]

I had 2 paired end data.Just 2. i know it is not usual. Those data were belong to one of my partners, those were all data she had. this was the process: 1:cutting adapters 2: Map with Star 3: Cout with Feature Count 4: Differentiated with Deseq2 Then i have 15 gene correctly differentiated and related to her subject which was Drug resistance.

[Kevin Blighe]

So, you performed PCA using just the 15 genes? You should generate a PCA bi-plot using all genes. Using the 15 genes, I would generate a heatmap.

[sabaghianamir70]

No. you get it wrong. i had more. but 15 of them had the best pvalue and fold changes. and 10 of them were had very good related to my subject of interest. i know it is un usual. but this is what i want to know. Why even this results supposed to be wrong, it seems like a logical answer

[swbarnes2]

Your p-values are meaningless, and your fold changes might be nonsense, because you have no replicates.

[sabaghianamir70]

I got results, i got 15 Gene differentiated and those genes are related to my subject which was Drug resistance.

[Mensur Dlakic]

As a general rule in this type of analysis, I would be more inclined to worry about your PC1 explaining 98% of variance rather than PC2 being at 2%. Setting some exotic explanations aside, this means that 1 or 2 of your features explain almost all differences between the datasets. Maybe it will be more clear when you explain the procedure further, but in my experience PC1 is rarely at 98% or even close.

Answer 1

[swbarnes2]

i had some error for analyzing them and the results were wrong. So i turned them into single ends, and i got acceptable results.

That was a really really horrible idea. You have no replicates, so you are lying to your software pretending you have replicates. When you lie to software, you usually get garbage as a result.

You might have gotten results, but they are not 'acceptable'. The fold changes are likely correctly showing you the difference between your samples (though with no replicates its impossible to know how robust those value are), but the p-values are trash.

i worried about this plot because of that 2%.

I don't think you should expect that read 1 and read 2 will yield identical results. That's where your 2% is coming from.

[sabaghianamir70]

Im understand that. But this data was all thing my partner had. In this situation, when i dont have any solution, this was my only hope, which the result was pretty good. and even related to her subject. In other papers, those genes were related to drug resistance which was her subject.