Issue with FEELnc_codpot spethres option.

[ttttuck]

Hello,

How are you?

I'm facing an issue when running the FEELnc_codpot module to find new human noncoding transcripts. I'm running the following code:

FEELnc_codpot.pl -i humanPutativeLncRNAs.gtf -a gencode.v41.pc_transcripts.fa -g GRCh38.primary_assembly.genome.fa -l gencode.v41.lncRNA_transcripts.fa -n 5000,5000 --spethres=0.85,0.85

And I'm not getting the 85% Se/Sp TG-ROC and two cutoffs. The program is providing only the 95% Se/Sp and one cutoff stating that this is the best value. I know it is but I want to see how it goes when I set the Se/Sp to 85%.

WARNING: the value where sensitivity equal specicifity: '0.9499412' is greater than the specificity threshold: mRNA: '0.85'; lncRNA: '0.85'. Use the best value.

WARNING: the threshold obtain for mRNA '0.389' is lesser than the the one for lncRNA '0.389'. Use the threshold '0.389' where sensitivity equal to specicifity for mRNA: '0.9499412'.

How can I "force" the code to provide me a 85% Se/Sp TG-ROC curve and two cutoffs?

Thanks in advance

Tlucky

[vwucher]

Hi @TluckDucky,

Apparently it seems that the issue here is that, since the two values you provided (0.85) are lower than the value obtain with the best threshold (0.95), i.e. when Se==Sp, then we force it to have only one threshold. Because here, if FEELnc used the thresholds obtained for 0.85 Se/Sp, then you will have some transcripts that will be both mRNA and lncRNA. That's why we decided to enforce this. If you want to get them anyway, you can modify codpot_randomforest_2thres.r and write the S and P R objects, they should contain the sensitivity and specificity of the transcripts as well as their names/ids (or others but normally it should be them). Then you can put manually the thresholds you want.

Hope this help! Bye, Valentin