Ballgown R Package

Question

[HaroonPakistan]

Hi seniors! First of all, I am very thankful to you for being here.!

I am using a ballgown for the DEG analysis. I have two samples (b73,f1). I have no replicate. Both samples are different from each other. I wanted to do a stattest and used this command;

Command; results_transcripts <- stattest(bg_filt, feature="transcript", covariate="sex", adjustvars = c("population"), getFC=TRUE, meas="FPKM")

Output:

Error in stattest(bg_filt, feature = "transcript", covariate = "sex", : There must be at least two replicates per group. Make sure covariate is categorical; if continuous, consider the timecourse option, or specify your own models with mod and mod0.

So, I thought maybe I have not enough samples to run Ballgown for the Statest. I just copied the samples and created two new samples which mean, Now I have 4 samples. And I run the same command as shown in above. Now, I got this error:

output: Coefficients not estimable: libadjust Error in solve.default(t(mod) %*% mod) : system is computationally singular: reciprocal condition number = 1.06159e-26 In addition: Warning message: Partial NA coefficients for 19766 probe(s)

Request: I am new in this field and really want to learn. Could you please help me? Anyone?

[cpad0112]

if you want to learn RNA seq data analysis, you can follow well laid out documentation like here and you can download data from here. Please make sure that you use/simulate data with discretion. Otherwise, it would result in output like this. Please consult local statistician to understand the statistical operations in high dimensional data analysis.

Answer 1

[swbarnes2]

The good news is, with only two samples, you can't really use any clever software. Just put your counts in Excel. I don't know ballgown, but my guess is it's freaking out because your 'replicates' have zero variance. And come on. You have two samples. You can't possibly adjust for a covariant. It's math, not magic.

[HaroonPakistan]

Ok. But my supervisor gave only three samples? He said you can just put another file by extracting random sequences from the same file. Thus it would become a replicate.

[Ram]

This is a toy example for learning and not a real scientific experiment, correct? If it is supposed to be a real experiment, faking a replicate is a bad choice.

[swbarnes2]

We'll, you can do a lot of things...but just because you can doesn't mean your results will mean anything. GIGO. If you put in fake data, expect fake results.