We ran our iPSC-cardiomyocyte model against the 28-drug CiPA panel. Here are two of those runs in full: dofetilide, where the high-risk call is textbook, and verapamil, where two channels fight each other and the honest answer is messier.
context
the validation setup
A plain Kernik-Clancy 2019 iPSC-CM action-potential model, scored against the canonical 28-drug CiPA reference. Drug effect enters as a static Hill block at 1 to 4x free Cmax, on the channels the panel shares: hERG, ICaL, INa-peak.
classifier
high-vs-not AUC (28)
low-vs-not AUC (28)
iPSC-CM multichannel
0.91
0.85
hERG-only baseline
0.89
0.77
published ORd/qNet (validation set)
1.00
0.89
"On the standard healthy-cell CiPA panel, a plain Kernik iPSC-CM AP model separates high-risk drugs at AUC ~0.91, competitive but below the adult-ventricular ORd/qNet gold standard and not better than a hERG-only margin on the high-risk call."
In other words, the iPSC model holds its own on the high-risk call. ORd/qNet still beats it there, and a hERG-only margin matches it. Where multichannel information earns its keep is the safe-drug call. Our own verdict on this run: "This is a respectable baseline, not a classification record."
case 1 · dofetilide
the ground truth
CiPA class: High (training set; the panel carries 8 high-risk drugs).
input · channel-panel IC50s · free Cmax 2.0 nM
current
IC50 (nM)
Hill
hERG (static fit)
4.87
0.93
ICaL
260
1.16
INa-peak
380
0.89
IKs was not measured for dofetilide.
output · predicted class and score
iPSC-CM multichannel
High
risk score 2.0
hERG-only baseline
High
score -0.386
Both models call it High, which is what the field expects from this drug. The 2.0 score is a sentinel: at this level of hERG block the model cell stops repolarizing, and the score is pinned at the ceiling.
the mechanism
Textbook hERG blocker. The static IC50 of 4.87 nM sits well below the 2.0 nM free Cmax, so therapeutic concentrations block most of the current.
case 2 · verapamil · the hard multi-channel case
the ground truth
CiPA class: Low (training set).
input · channel-panel IC50s · free Cmax 81 nM
current
IC50 (nM)
Hill
hERG (static fit)
288
0.96
ICaL
202
1.1
The reference panel also measured INaL (IC50 7.03e+03 nM, Hill 1.03), but the model has no distinct INaL current, so that number went nowhere. INa-peak was never measured for verapamil. What the model actually saw was hERG and ICaL block.
output · predicted class and score
iPSC-CM multichannel
Intermediate
risk score 0.19
hERG-only baseline
High
score -0.551
The truth is Low, two classes below the hERG call. The multichannel model lands at Intermediate, one class away. And the hERG baseline is the only training-set Low that any classifier called High.
reading the result
Verapamil blocks hERG and Cav1.2. Calcium block offsets hERG block and limits repolarization delay. The CiPA literature calls verapamil a balanced ion channel blocker with low TdP risk for exactly this reason (Vicente et al. 2017, doi:10.1002/cpt.896). A hERG-only margin sees half the picture, so it calls verapamil High. Reading both channels pulls the call back toward the truth.
The model still lands one class high, and we do not have a clean story for why. Our validation runs did not isolate the cause, and the validation record says nothing about verapamil by name. Rather than guess, we are reporting the miss as a miss.
The pattern across the whole panel is at least the kind of mistake you can live with. "Adjacent accuracy is perfect (1.00): the model never calls a Low drug High or vice versa, so failures are conservative, not dangerous."
known limits, from our validation report
No late sodium (INaL). Kernik-2019 has no distinct INaL current, so INaL block is unrepresentable. INaL-blocking safe drugs lose their protective signal.
Static hERG block. The applied block is a static pore block. The state-dependent hERG binding model behind ORd/qNet's near-perfect result does not run here.
Healthy-cell substrate. A healthy-cell iPSC-CM model on a healthy-cell benchmark plays on the incumbent's home field. No patient or variant biology is in play.
where this leaves things
So where does that leave the model? It runs a recognized benchmark and holds its own: AUC 0.91 on the high-risk call, and a real edge over the hERG-only margin in spotting safe drugs (0.85 against 0.77). ORd/qNet still wins the healthy-cell panel outright, and we are not claiming otherwise. As our own validation report puts it, "ORd/qNet already classifies the healthy-cell CiPA panel near-perfectly. Rebuilding that with an iPSC-CM model is not a moat."
Two caveats worth knowing before you weigh these numbers. Both drugs above come from the 12-compound training set, so the risk thresholds were fit with them in view. And a healthy-cell benchmark says nothing about how the model would read your compound, which is the actual product: a mechanism read on your flagged molecule, on this same validated substrate. If that is the question you have, write to us and we will run it.
AUC, risk scores and classes from the CiPA validation record (T-151, 2026-07-03). Per-drug IC50, Hill and Cmax values from the CiPA 28-drug reference dataset (T-150).