How do I obtain logFC values from sleuth results (both genes and isoforms)?

Question

[cpcantalapiedra]

I have run sleuth (0.28.0) for the first time, and I have results both for isoforms and for genes. The results from isoforms were obtained with the R function "sleuth_results", yielding a header and a sample line like:

"target_id" "pval"  "qval"  "b" "se_b"  "mean_obs"  "var_obs"   "tech_var"  "sigma_sq"  "smooth_sigma_sq"   "final_sigma_sq"    "ens_gene"  "ext_gene"
"1" "comp53082_c1_seq64"    1.51049897111248e-63    4.0609764838359e-59 -7.63767780030034   0.453855586263854   3.12569171959022    19.5061503349294    0.0380559668566394  0.0541084449372053  0.167928926326268   0.167928926326268   "comp53082_c1"  "comp53082_c1"

The results from genes were obtained with the R function sleuth_gene_table, yielding a header and a sample line like:

"ens_gene"  "most_sig_transcript"   "pval"  "qval"  "num_transcripts"   "list_of_transcripts"
"1" "comp12308_c0"  "comp12308_c0_seq1" 1.18666532051623e-21    4.55764244886839e-18    1   "comp12308_c0_seq1"

I would like to obtain logFC values or at least be able to say which genes are "up-regulated" and which ones "down-regulated".

• From isoforms data maybe I could use the "b" (beta" value). Does positive "b" mean up or down regulation?
• Is there any way to obtain logFC from genes data?

Answer 1

[andrew.j.skelton73]

You're looking for 'b', as you suspected. I've been interpreting it similar to a fold change. The manual states:

b: 'beta' value (effect size). Technically a biased estimator of the fold change

So greater than zero, positive fold change, less than zero, negative fold change.

[cpcantalapiedra]

Thank you. And what about genes data?

[andrew.j.skelton73]

For Gene level stuff I've been using DESeq2, with the tximport function. DESeq2 appears to be much much more flexible in the design of contrasts over sleuth.

[cpcantalapiedra]

uumm got no much experience with DESeq2. Contrasts can be designed in sleuth with the common GLM syntax of R within a matrix data type. For genes I got in sleuth the p and qvalues of the contrasts, so I guessed that the software had calculated somehow a beta and therefore a logFC should be available to be obtained, but I did not find it.

[andrew.j.skelton73]

Contrasts can be designed in sleuth with the common GLM syntax of R within a matrix data type

Could you expand, and maybe show some code? Feel free to amend your post with the code you used to produce your results. There's currently a lot of requests for ways to perform pairwise contrasts with more than 2 groups, and non trivial model designs, so if you have a way, I think it'd benefit a lot of users to have a worked example.