Difference of log2 values between .cnr file and .bintest

[Fponelle]

Hi,

Why log2 values are different between cnr file and bintest.cns file?

Here is my example:

- In .cnr file, all bins on BRCA2 have a log2 between 1 and 2.5 suggesting and amplification of the entire gene:

chromosome	start	end	gene	log2	depth	weight
chr13	32890559	32890670	BRCA2_ex02	1.92172	11112.5	0.989018
chr13	32893193	32893468	BRCA2_ex03	2.06781	9570	0.989823
chr13	32899192	32899327	BRCA2_ex04	1.29903	4212.33	0.989634
chr13	32900217	32900293	BRCA2_ex05	1.04573	1308.66	0.98301
…
chr13	32968805	32969076	BRCA2_ex25	1.86698	12887.1	0.993064
chr13	32971014	32971187	BRCA2_ex26	1.72835	12128.8	0.991547
chr13	32972278	32972595	BRCA2_ex27	1.55005	14927.4	0.993474
chr13	32972595	32972913	BRCA2_ex27	1.88532	14994.7	0.993332

- .cns file agrees with .cnr file showing a log2 for the whole gene of 1.64:

chromosome	start	end	gene	log2	depth	probes	weight	ci_lo	ci_hi
chr13	30937699	32972913	BRCA2_ex02,BRCA2_ex03,BRCA2_ex04,BRCA2_ex05,BRCA2_ex06,BRCA2_ex07,BRCA2_ex08,BRCA2_ex09,BRCA2_ex10,BRCA2_ex11,BRCA2_ex13,BRCA2_ex14,BRCA2_ex15,BRCA2_ex16,BRCA2_ex17,BRCA2_ex18,BRCA2_ex19,BRCA2_ex20,BRCA2_ex21,BRCA2_ex22,BRCA2_ex23,BRCA2_ex24,BRCA2_ex25,BRCA2_ex26,BRCA2_ex27	1.63355	7483.63	60	58.3449	1.5957	1.66772

- call.cns reports a CNV on the BRCA2 gene by counting 7 copies with a mean log2 of 1.6:

chromosome	start	end	gene	log2	cn	depth	p_ttest	probes	weight
chr13	30937699	32972913	BRCA2_ex02_UTR,BRCA2_ex02,BRCA2_ex03,BRCA2_ex04,BRCA2_ex05,BRCA2_ex06,BRCA2_ex07,BRCA2_ex08,BRCA2_ex09,BRCA2_ex10,BRCA2_ex11,BRCA2_ex13,BRCA2_ex14,BRCA2_ex15,BRCA2_ex16,BRCA2_ex17,BRCA2_ex18,BRCA2_ex19,BRCA2_ex20,BRCA2_ex21,BRCA2_ex22,BRCA2_ex23,BRCA2_ex24,BRCA2_ex25,BRCA2_ex26,BRCA2_ex27	1.60607	7	7483.63	1.29963e-39	60	58.3449

- But in bintest.cns file, for the same bins, log2 are really different, some of them are negative: (Initialy those bins were not reported in bintest file, I modified alpha value to 1 in bintest.py to understand why my bins were not reported as significative.)

chr13	32890559	32890670	BRCA2_ex02	0.288171	11112.5	0.989018	0.0161461
chr13	32893193	32893468	BRCA2_ex03	0.434261	9570	0.989823	8.8159e-05
chr13	32899192	32899327	BRCA2_ex04	-0.334525	4212.33	0.989634	0.00354805
chr13	32900217	32900293	BRCA2_ex05	-0.587826	1308.66	0.98301	3.62753e-05
…
chr13	32968805	32969076	BRCA2_ex25	0.233426	12887.1	0.993064	0.014072
chr13	32971014	32971187	BRCA2_ex26	0.0948008	12128.8	0.991547	0.411222
chr13	32972278	32972595	BRCA2_ex27	-0.0835012	14927.4	0.993474	0.411222
chr13	32972595	32972913	BRCA2_ex27	0.251767	14994.7	0.993332	0.00656489

Can you help me understand why there is such a difference?

Thanks

Flora

[etal]

Is this happening with CNVkit version 0.9.8?

[Fponelle]

I'm sorry I forgot to specify the version but yes I am using 0.9.8

[Fponelle]

Hi @etal , I would like to contact you again to find out if you have found some answers to my problem?

[tskir]

Hi @Fponelle, looking at the code, it looks like the log2 values in the bintest output are residuals: https://github.com/etal/cnvkit/blob/25d91cdb962639c98d041c3b05c7de659c41dc70/cnvlib/bintest.py#L26

If the segments are provided, the residuals are calculated compared to the segment to which the bin belongs, and otherwise to the whole chromosome: https://github.com/etal/cnvkit/blob/25d91cdb962639c98d041c3b05c7de659c41dc70/cnvlib/cnary.py#L417-L429

I suspect that you're running bintest with the segments specified. And so, given that the whole BRCA2 is duplicated, the change of CN of any given bin is not relevant compared to the BRCA2 segment.

Could you try re-running without the -s parameter to see if this fixes your issue?

In any case, it looks like documentation on bintest could be improved to reflect this behaviour.

[Fponelle]

Hi @tskir Thank you for your answer, my problem is solved. By default in the "batch" command the segments are specified, you must launch the "bintest" command separately to be able to remove this parameter.

[etal]

Yes, that's right. With -s, you're telling bintest that those are the alterations you already know about, and asking for any additional bin-level alterations. Maybe that subtraction should be used only for the reported p-values, and not alter the log2 values in the output.

[tskir]

Note to self — Leaving this issue open to improve bintest documentation in the future

[tskir]

Help wanted: for someone to document bintest and add the section to the relevant part of CNVkit docs