CGI

Install packaged CpG-island BEDs or download fresh UCSC cpgIslandExt tables:

sjcab-peak2anno-db install-cgi
sjcab-peak2anno-db install cgi
sjcab-peak2anno-db download-cgi --species hg38 -d cgi_downloads

Outputs are stored as {db-path}/cgi/{species}_cgi.bed.

Resource URLs

Liftover comparison

The following is an interval-overlap comparison of the generated liftover files against native target CGI BED files:

  • The native intervals are the true/reference set; liftover intervals are the predictions being evaluated.
  • TP: native truth intervals with at least one base overlapping a liftover interval.
  • FN: native truth intervals without an overlap (native truth - TP).
  • TPR = TP / native truth; FNR = FN / native truth.
  • merge is applied to the liftover result before overlap testing. none means sorted but unmerged input; the other values are bedtools merge -d N.
assembly liftover merge native truth liftover intervals TP TPR FN FNR
hg19 CrossMap none 30,344 33,643 27,183 89.583% 3,161 10.417%
hg19 CrossMap 100 30,344 27,777 27,183 89.583% 3,161 10.417%
hg19 CrossMap 200 30,344 27,555 27,183 89.583% 3,161 10.417%
hg19 UCSC liftOver none 30,344 29,997 27,112 89.349% 3,232 10.651%
hg19 UCSC liftOver 100 30,344 27,605 27,112 89.349% 3,232 10.651%
hg19 UCSC liftOver 200 30,344 27,411 27,112 89.349% 3,232 10.651%
mm10 CrossMap none 17,017 479,438 13,735 80.713% 3,282 19.287%
mm10 CrossMap 100 17,017 23,535 13,735 80.713% 3,282 19.287%
mm10 CrossMap 200 17,017 22,755 13,735 80.713% 3,282 19.287%
mm10 UCSC liftOver none 17,017 5,272 2,523 14.826% 14,494 85.174%
mm10 UCSC liftOver 100 17,017 4,717 2,523 14.826% 14,494 85.174%
mm10 UCSC liftOver 200 17,017 4,698 2,523 14.826% 14,494 85.174%

Prediction-side overlap

If the question is “how many liftover regions overlap native truth?”, the corresponding metric is prediction precision/positive predictive value (PPV), not TPR. Here liftover TP is a liftover interval overlapping at least one native truth interval, and liftover FP is a liftover interval with no native overlap. The native-truth recall columns are retained for reference.

assembly liftover merge liftover intervals liftover TP PPV liftover FP native TP native TPR
hg19 CrossMap none 33,643 32,329 96.094% 1,314 27,183 89.583%
hg19 CrossMap 100 27,777 27,037 97.336% 740 27,183 89.583%
hg19 CrossMap 200 27,555 26,835 97.387% 720 27,183 89.583%
hg19 UCSC liftOver none 29,997 29,283 97.620% 714 27,112 89.349%
hg19 UCSC liftOver 100 27,605 26,934 97.569% 671 27,112 89.349%
hg19 UCSC liftOver 200 27,411 26,748 97.581% 663 27,112 89.349%
mm10 CrossMap none 479,438 251,575 52.473% 227,863 13,735 80.713%
mm10 CrossMap 100 23,535 13,392 56.903% 10,143 13,735 80.713%
mm10 CrossMap 200 22,755 13,222 58.106% 9,533 13,735 80.713%
mm10 UCSC liftOver none 5,272 2,734 51.859% 2,538 2,523 14.826%
mm10 UCSC liftOver 100 4,717 2,414 51.177% 2,303 2,523 14.826%
mm10 UCSC liftOver 200 4,698 2,409 51.277% 2,289 2,523 14.826%

Confidence and limitations

These figures are overlap-based recall only. They do not measure false positive intervals, coordinate accuracy, one-to-many mappings, or preservation of CGI metadata. The very large CrossMap interval counts, especially for mm10, indicate fragmentation or one-to-many mappings and should not be interpreted as higher confidence. UCSC liftOver recovers only 14.826% of the mm10 native intervals in this comparison, so that result should not be used without review against a native mm10 CGI release. Results can change with the selected chain file, source release, chromosome naming, and merge policy.