b86eab362ce697c350764ffe438ccdba8829e0f3
gperez2
  Mon Jul 20 17:37:35 2026 -0700
Fixing the NMDetective-A/B score-range text (0 to 1, not -1 to +1), updating the citation to Lindeboom et al. 2019, and changing viewLimits to -0.3:1.5 on all four NMDetective subtracks. refs #37843

diff --git src/hg/makeDb/trackDb/human/hg38/nmd.html src/hg/makeDb/trackDb/human/hg38/nmd.html
index 58e22b0acce..0747c3cea4e 100644
--- src/hg/makeDb/trackDb/human/hg38/nmd.html
+++ src/hg/makeDb/trackDb/human/hg38/nmd.html
@@ -28,34 +28,39 @@
     (Plus Clinical).</li>
   <li><b><a href="hgTrackUi?g=nmdEscGencode">NMD escape Gencode</a></b>:
     NMD escape regions derived from GENCODE V49 transcripts.</li>
   <li><b><a href="hgTrackUi?g=nmdEscNcbiRefSeq">NMD escape NCBI RefSeq</a></b>:
     NMD escape regions derived from NCBI RefSeq Curated transcripts
     (NM_ and NR_ accessions only).</li>
 </ul>
 <p>
 Click either of the links to the track details here or above to show the four rules
 that were used (50 bp, intronless, 100 bp, long exon &gt;400 nt).
 </p>
 
 <h3>NMDetective scores</h3>
 <p>
 Machine-learning predictions of NMD efficiency from
+<a href="https://www.ncbi.nlm.nih.gov/pubmed/31659324" target="_blank">Lindeboom
+et al. 2019</a> (NMDetective-A and NMDetective-B models, trained on
 <a href="https://www.ncbi.nlm.nih.gov/pubmed/27618451" target="_blank">Lindeboom
-et al. 2016</a> (A and B models) and from Veiner <em>et al.</em>
-(NMDetective-AI, pre-print 2026). Positive scores indicate predicted NMD
-triggering; negative scores indicate predicted escape.
+et al. 2016</a>) and from Veiner <em>et al.</em>
+(NMDetective-AI, pre-print 2026). NMDetective-A and NMDetective-B scores range
+from 0 to 1, with values near 1 indicating predicted NMD triggering and values
+near 0 indicating predicted escape. NMDetective-AI uses a different scale,
+roughly -1.1 to +1.5, with higher values indicating triggering and lower
+values indicating escape.
 </p>
 <ul>
   <li><b><a href="hgTrackUi?g=nmdDetectiveA">NMDetective-A</a></b>:
     Random forest model for all possible PTCs from nonsense variants.</li>
   <li><b><a href="hgTrackUi?g=nmdDetectiveB">NMDetective-B</a></b>:
     Decision tree model for all possible PTCs from nonsense variants.</li>
   <li><b><a href="hgTrackUi?g=nmdDetectiveA_ptc">NMDetective-A PTC</a></b>:
     Random forest model for the first out-of-frame PTC from frameshifting indels.</li>
   <li><b><a href="hgTrackUi?g=nmdDetectiveB_ptc">NMDetective-B PTC</a></b>:
     Decision tree model for the first out-of-frame PTC from frameshifting indels.</li>
   <li><b><a href="hgTrackUi?g=nmdDetectiveAi">NMDetective-AI</a></b> and
     <b><a href="hgTrackUi?g=nmdDetectiveAiBed">NMDetective-AI variants</a></b>:
     Deep-learning model on MANE Select transcripts (GENCODE V46). Signal track
     shows the position-averaged prediction; variants track shows one item per
     stop-gain mutation per codon.</li>