Verapamil blocks hERG potently and never causes TdP in the clinic, because it blocks calcium at the same time and the two effects cancel. hERG-only screening calls it High. The mechanism read calls it Intermediate. It is Low, and the miss is on the page.
If a single hERG margin were enough, verapamil would be a High-risk drug. It is a potent hERG blocker, its free margin sits under the boundary that triggers a High call, and it is prescribed as a calcium-channel blocker without a torsades problem in the clinic. The margin is not wrong because the data are wrong. It is wrong because it reads one channel and the drug works on two.
The two blocks that cancel
Verapamil blocks the delayed rectifier hERG/IKr and the L-type calcium current ICaL at exposures within a factor of 1.4 of each other. The hERG block lengthens repolarization. The calcium block shortens it. In the model the two nearly cancel: at 1x free Cmax the read reports a +5.7% APD90 change, against dofetilide's +15.3% at its own 1x.
input data · CiPA reference panel
current
IC50 (nM)
Hill
block at 1x Cmax
in the scored model
Cav1.2 / ICaL
202
1.10
26.8%
yes
hERG / IKr
288
0.96
22.8%
yes
late Na / INaL
7.03e3
1.03
1.0%
no
Ito
1.34e4
0.82
1.5%
no
Kir2.1 / IK1
3.49e8
0.27
<0.1%
no
Nav1.5 / INa peak, IKs
no values in the public panel
Free Cmax 81 nM. The two IC50s are within a factor of 1.4 of each other.
Separation between dofetilide and verapamil is produced entirely by the calcium block, and it is invisible to a method that reads only hERG. At 4x free Cmax, dofetilide has stopped repolarizing while verapamil has prolonged by 19%. The margin sees two High-risk drugs. The market sees one withdrawn and one safe. The difference is a second current.
The limitation is structural, and now it is confirmed in a third independent place. Verapamil's calcium block is the one case where the model's direction disagrees with Lee et al. 2025. The validation that names this is published in full.
the read for a multichannel compound
A compound that acts on more than one channel is the rule. If your programme reads the panel you already have, the read returns which currents matter and the one experiment that confirms it. The first read is free. What a free read covers.
Ion-channel IC50, Hill and free Cmax from the CiPA 28-drug reference dataset (FDA/CiPA public repository, branch Model-Validation-2018; Li et al. 2017); TdP class from Colatsky et al. 2016. Model runs on the Kernik-Clancy 2019 human iPSC-CM model (doi.org/10.1113/JP277724) integrated with Myokit/CVODE, frozen CiPA validation run T-151 (2026-07-03). Mechanism ranking unvalidated against wet data.