How to calculate FPR from dwgsim_eval output?

[yjx1217]

Hello,

I am using dwgsim_eval to evaluate different mapper's performance on my simulated reads. I want to plot a ROC graph about the result, so I need to calculate True Positive Rate (TPR) and False Positive Rate (FPR) for each mapping quality threshold. I guess TPR should be column 17 ("sensitivity
# (mc' / (mc' + mi' + mu')) | sensitivity: the fraction of reads that should be mapped that are mapped correctly at or greater than the threshold"), then how should I calculate FPR? Thanks!

BTW:
I am quite confused with the following 3 definitions:
# mu | the number of reads unmapped that should be mapped be mapped at the threshold
# um | the number of reads mapped that should be unmapped be mapped at the threshold
# uu | the number of reads unmapped that should be unmapped be mapped at the threshold

[RockChalkJayhawk]

You can find your answer here: http://en.wikipedia.org/wiki/Sensiti...nd_specificity

[yjx1217]

You can find your answer here: http://en.wikipedia.org/wiki/Sensiti...nd_specificity

Thanks RockChalkJayhawk! I've checked the definition of FPR(=FP/(FP+TN)) but I think my problem is that I am not sure how to find the corresponding term of FP and TN from the output of dwgsim_eval.

The list of dwgsim_eval is given as follows:

# thr | the minimum mapping quality threshold
# mc | the number of reads mapped correctly that should be mapped at the threshold
# mi | the number of reads mapped incorrectly that should be mapped be mapped at the threshold
# mu | the number of reads unmapped that should be mapped be mapped at the threshold
# um | the number of reads mapped that should be unmapped be mapped at the threshold
# uu | the number of reads unmapped that should be unmapped be mapped at the threshold
# mc' + mi' + mu' + um' + uu' | the total number of reads mapped at the threshold
# mc' | the number of reads mapped correctly that should be mapped at or greater than that threshold
# mi' | the number of reads mapped incorrectly that should be mapped be mapped at or greater than that threshold
# mu' | the number of reads unmapped that should be mapped be mapped at or greater than that threshold
# um' | the number of reads mapped that should be unmapped be mapped at or greater than that threshold
# uu' | the number of reads unmapped that should be unmapped be mapped at or greater than that threshold
# mc' + mi' + mu' + um' + uu' | the total number of reads mapped at or greater than the threshold
# (mc / (mc' + mi' + mu')) | sensitivity: the fraction of reads that should be mapped that are mapped correctly at the threshold
# (mc / mc' + mi') | positive predictive value: the fraction of mapped reads that are mapped correctly at the threshold
# (um / (um' + uu')) | false discovery rate: the fraction of random reads that are mapped at the threshold
# (mc' / (mc' + mi' + mu')) | sensitivity: the fraction of reads that should be mapped that are mapped correctly at or greater than the threshold
# (mc' / mc' + mi') | positive predictive value: the fraction of mapped reads that are mapped correctly at or greater than the threshold
# (um' / (um' + uu')) | false discovery rate: the fraction of random reads that are mapped at or greater than the threshold

[nilshomer]

See the discussion here:

http://sourceforge.net/mailarchive/forum.php?thread_name=CAEFx-Fz-u-4yRJ4QEhioQ_HyhNJSSFtrzqQEPV%3D1cNR_NM%2Bb_Q%40mail.gmail.com&forum_name=dnaa-help

What is a false positive here though, since we could have a read that is mapped and it can be "wrong" if:
#1 mapped to the wrong position
#2 not mapped

#2 would seem to be a FP, but #1 doesn't fit into the FP/TP/FN/TN scheme. Hence use positive predictive value, since it only hurts sensitivity when a read does not map (so only care if it does map). Use a ROC plotting sensitivity vs. PPV.