70ec78b7ef15cc060932b5b8dd436fd34bcbdaf1
jnavarr5
  Fri Sep 25 13:31:22 2026 -0700
Adding a paragraph about a filtering issue that happens when there is a tie in the absolute max value. refs #38112

diff --git src/hg/makeDb/trackDb/human/hg38/epigenCentral.html src/hg/makeDb/trackDb/human/hg38/epigenCentral.html
index c1c8455e795..38afcd2ae9c 100644
--- src/hg/makeDb/trackDb/human/hg38/epigenCentral.html
+++ src/hg/makeDb/trackDb/human/hg38/epigenCentral.html
@@ -22,30 +22,37 @@
 <p>
 Each CpG site is drawn as a single-base feature, colored by the direction of the methylation
 change. Delta-beta is the mean methylation in cases minus the mean methylation in controls, so a
 positive value is a gain of methylation in cases and a negative value is a loss. Where a site
 belongs to several episignatures, the color, the mouse-over and the delta-beta shown come from
 the episignature with the largest absolute delta-beta at that site.
 </p>
 
 <table class="stdTbl">
   <tr><th style="background-color:#FF5F01;width:2em">&nbsp;</th>
       <td>Gain of methylation in cases, positive delta-beta</td></tr>
   <tr><th style="background-color:#0072EB;width:2em">&nbsp;</th>
       <td>Loss of methylation in cases, negative delta-beta</td></tr>
 </table>
 
+<p>
+The <b>Direction of methylation change</b> filter uses the same strongest episignature, so it does
+not catch a site where a weaker episignature changes in the opposite direction; about 1,600 of
+the 2,850 sites with more than one episignature are of this kind. The table on the details page
+lists every episignature at the site with its own direction.
+</p>
+
 <p>
 The mouse-over gives the probe ID, how many episignatures cover the site, and the episignature
 with the largest effect there with its disorder, direction and delta-beta.
 </p>
 
 <p>
 Clicking a site opens a page with a table of every episignature at that site, one row each,
 giving the gene or locus, the disorder, its OMIM entry, the direction, the delta-beta, the
 adjusted p-value and the multiple-testing correction that the publishing study used. Studies
 differ in which correction they applied, so the adjusted p-values in one row are not always
 directly comparable with those in another.
 </p>
 
 <p>
 Three filters are available on the track configuration page. The episignature filter keeps only