You send a standard ion-channel panel and a free Cmax. The read returns ranked ion-current mechanisms, a TdP-risk class, a dose-response, and the one experiment that would confirm the cause. It also states, in writing, the scope of each part.
The read is two runs of the same model. The forward half applies static Hill block to the model's conductances at 1x, 2x, 3x and 4x free Cmax, simulates the action potential, and reports APD90. A frozen pair of thresholds turns the largest fractional change into one of three TdP classes. The inverse half throws away the known panel, takes only the resulting action-potential shape, and searches 1,560 conductance hypotheses over six currents for the ones that reproduce that shape, then names which hypotheses it can tell apart.
What comes back
A risk class. Low, Intermediate or High, from a frozen pair of APD90 thresholds.
Ranked mechanisms. Which conductance changes reproduce the signal, with the gap between ranks stated. When nothing is identifiable, the report says so.
A dose-response. The block and the APD90 change at 1x to 4x free Cmax.
A confirming experiment. One assay, chosen to kill the surviving ambiguity.
Stated scope. Which inputs were used, how well each current resolves, and where the report says "unclear".
the risk score uses three currents
The three-class score is driven by hERG/IKr, Cav1.2/ICaL and peak Nav1.5/INa. IKs, IK1, Ito and late sodium are read and reported; the scored simulation uses the first three. The inverse search is wider, over six currents. Late sodium stays out of both, because the Kernik-Clancy formulation carries no distinct INaL current. That absence decides some cases.
It is a single cell. It bounds a cellular cause; conduction, QRS and re-entry live in the whole-heart models downstream of it.
It inherits the model's gaps. A compound whose margin depends on late-sodium block will read too dangerous, and we flag that before you buy. The ranolazine case is the stated example.
Missing inputs widen the answer. Four of the eight panel currents may be unreported for your compound. The read still runs; the width is reported.
stated limits
We publish the benchmark we lose. On the 28-drug CiPA panel a plain hERG margin gets 18 of 28 right and we get 17. The loss is on the site.
You hear about a compromised read before you pay. If your compound's measured INaL IC50 sits within about tenfold of its hERG IC50, the model's missing INaL current costs the read resolving power, and we say so in the quote.
see it on real compounds
Four public compounds are read end to end on this site, with two wrong answers shown in full. Read them. Then send one compound of yours, free, for the same report format. What a free read covers.
Pipeline as run for the reads page and the frozen CiPA validation (T-151, 2026-07-03). Model: Kernik-Clancy 2019 human iPSC-CM (doi.org/10.1113/JP277724) integrated with Myokit/CVODE.