The same report you would later pay for, on a compound you are free to share. Below is one, section by section, on verapamil, with a note under each section explaining what it is for and where it can mislead you. Nothing is held back to make the paid version look better.
In-silico safety work has a trust problem it earned. The output is a number produced by a model you cannot inspect, delivered by someone who is paid whether or not it is right. No amount of description fixes that. The only useful answer is to hand over a complete piece of work and let you judge it.
The free read is the identical report format, run on a compound where the confidentiality problem does not arise. It is never invoiced and there is no obligation attached to it. If you read it and conclude the method is not for you, that outcome is worth more to us than a customer who bought on a brochure.
what we need from you
One compound. Published, or otherwise non-confidential. A marketed drug, a tool compound, a competitor's disclosed asset, or your own compound if you are free to share the channel data.
Ion-channel data. IC50 and Hill coefficient where you have them, or percentage block at named concentrations. Three currents drive the risk score: hERG/IKr, Cav1.2/ICaL and peak Nav1.5/INa. Values for IKs, IK1, Ito and late sodium are useful context and are recorded in the report, but they do not enter the scored simulation. Say which currents you have and which you do not; a missing value is a stated width in the answer, not a reason to refuse the read.
An approximate free Cmax, if you have one. Total plasma concentration with a protein-binding estimate is fine. Without it we report block and APD90 against a concentration range instead of against exposure, and no risk class is assigned, because the class is defined relative to exposure.
One line of context. What decision this read feeds. It changes which ambiguity is worth resolving and therefore which confirming experiment we recommend.
That is the whole list. There is no intake form beyond the request page, no call required first, and no NDA needed for public data.
the sample report · verapamil, from public data
Verapamil is the right compound to show, because it is the one the method gets partly wrong. A sample report on an easy compound tells you nothing about how the hard parts are handled.
section 1 · inputs, restated so you can check them
current
IC50 (nM)
Hill
enters the simulation
Cav1.2 / ICaL
202
1.10
yes
hERG / IKr
288
0.96
yes
late Na / INaL
7.03e3
1.03
no, recorded only
Ito
1.34e4
0.82
no, recorded only
Kir2.1 / IK1
3.49e8
0.27
no, recorded only
Nav1.5 / INa peak
not supplied
IKs
not supplied
Free Cmax 81 nM. Model: Kernik-Clancy 2019 human iPSC-cardiomyocyte, drug-free APD90 413.31 ms in this run against 414 ms published.
Why this section exists. Every number downstream is a function of this table. If a value here is wrong or out of date, the report is wrong in a way you can trace in one step rather than argue about. It also states plainly that two of your seven currents were not simulated.
section 2 · ranked ion-current mechanisms
rank
hypothesis
fit
verdict
1
IKr 0.75 + ICaL 0.75
0.060
survives
2
IKr 0.90 + IK1 1.10
0.089
survives
3
IK1 1.10 + Ito 2.00
0.105
survives
-
best single-current answer (IK1 1.10)
0.171
rejected
Three hypotheses lie within a factor of 1.8 in fit error. Two of them disagree about whether the compound touches calcium at all. The data cannot separate them. The single-current search fails outright on this compound, which is itself a result: it means the mechanism is not one channel.
Why this section exists, and its main failure mode. A ranked list is only meaningful with the gaps attached. Rank 1 winning by a factor of three, as it does for dofetilide, is knowledge. Rank 1 winning by 0.0053, as it does for astemizole, is noise wearing a ranking's clothes. Any report that gives you an ordering without the separation is inviting you to over-read it. Ours states which entries survive.
section 3 · dose-response and APD90
multiple of free Cmax
IKr block
ICaL block
APD90
change
1x (81 nM)
22.8%
26.8%
436.7 ms
+5.7%
2x (162 nM)
36.5%
44.0%
453.8 ms
+9.8%
3x (243 nM)
45.9%
55.1%
470.9 ms
+13.9%
4x (324 nM)
52.8%
62.7%
490.9 ms
+18.8%
Why this section exists. The class in section 4 is a threshold applied to this curve, so the curve is the more informative object. It also shows the shape: verapamil prolongs steadily and never fails to repolarise, where dofetilide stops repolarising by 4x. If you only need one thing from a read, take this table.
section 4 · TdP-risk class
this read
Intermediate
score 0.19, driven by IKr against partial ICaL offset
published CiPA class
Low
we are one class too high
Thresholds were fitted once on twelve training compounds and frozen before the other sixteen were scored. They are not re-tuned per customer. Across all 28 this classifier is exactly right on 17 and never more than one class out.
Why this section exists, and why it is the least valuable one. A three-class label is what people ask for and the weakest thing we produce. On the public panel a plain hERG margin classifies slightly better than we do, and we publish that comparison. Buy the read for sections 2, 3 and 5. Treat section 4 as a cross-check.
section 5 · the confirming experiment
Record the cytosolic calcium transient on the same iPSC-CM well as the action potential, drug against paired vehicle.
surviving hypothesis
predicted Ca transient, % of control
IKr + ICaL block (rank 1)
88.6%
IK1 gain, no calcium block (rank 2)
103.2%
The surviving hypotheses move calcium in opposite directions, about 14 points apart, so the readout is a sign test rather than a magnitude test. It adds one dye to an assay you are already running. The 14 points are the model's prediction, not a measurement; a paired vehicle design and enough wells to resolve it are required.
Why this section exists. This is the part that changes what you do on Monday. The assay is chosen to kill the specific ambiguity left in section 2, which is why it differs per compound: verapamil gets a calcium readout, astemizole gets a direct IKs voltage clamp, ranolazine gets a late-sodium blocker add-back, and dofetilide gets told that no discriminating experiment is needed. If the surviving hypotheses cannot be separated by any assay we can name, the report says that instead of inventing one.
section 6 · limits of this read
The class is wrong by one step. Verapamil is clinically Low and this read says Intermediate.
Two supplied currents were not simulated, and two more were never measured. Peak sodium block is absent from the run entirely.
The model has no distinct late sodium current. Compounds whose margin depends on INaL block read too dangerous. Verapamil's INaL block is only 1% at 1x, so this one is not materially affected. Ranolazine is, and that read is published too.
Single cell. No conduction, no QRS, no re-entry. The read bounds a cellular cause; it does not predict an arrhythmia in a patient.
The mechanism ranking has never been validated against a measured trace. It is scored against phenotypes this model generated. It demonstrates resolving power and degeneracy; it is not a benchmarked prediction.
Why this section exists. It is the section a vendor has the most incentive to shorten, which is why it is the one to read first. Every limit above is either measured or structural. None of it is hedging.
what the free read does not include
Confidential compounds. The free read runs on published or shareable data only. That is what makes it free to give: we can publish the method applied to it. Proprietary structures and unpublished panels are paid work.
Confirming-experiment design for a proprietary compound. The free read names the assay class and the discriminating readout, as section 5 does above. Turning that into a protocol for your compound, with arms, concentrations, controls and a decision rule, is part of the paid read.
Wet-pilot protocol design. Specifying the iPSC-CM pilot you would commission at a CRO, including what each result rules out, is quoted separately and only after a read. We have not run a wet pilot for a customer.
An identifiability analysis on your own data. The paid read maps the conductance combinations your measured APD90 separates, and names the pairs it cannot. The free read shows the ranked list and the gaps, not the full map.
A regulatory submission, or advice about one. This is a mechanistic read. It is not a CiPA submission, it is not GLP, and it does not replace a thorough QT study.
turnaround
Days, for a free read on public or non-confidential data. The simulation itself is minutes. The time goes into checking the input panel against its source, running the hypothesis grid, and writing the limits section without trimming the parts that hurt.
A paid single-compound read on your own data takes one to two weeks and starts at $12,000. Those prices are being tested and may move; whatever is quoted before work starts is what you pay.
If your compound has a measured late-sodium IC50 within roughly a factor of ten of its hERG IC50, we will say up front that the read is compromised, and why, rather than deliver it and let you find out.
To request one, include: the compound name or structure; the ion-channel values you have, with units and the assay they came from; an approximate free Cmax if you have one; and one line on the decision this feeds. Email is fine if the form is not.
Verapamil ion-channel values and free Cmax from the CiPA 28-drug reference dataset, derived from the FDA/CiPA public repository and Li et al. 2017; TdP class from Colatsky et al. 2016. Action potentials are the Kernik-Clancy 2019 human iPSC-CM model (doi.org/10.1113/JP277724) integrated with Myokit/CVODE, re-run 2026-09-06. Risk class and thresholds are the frozen CiPA validation run (T-151, 2026-07-03). The 17-of-28 accuracy figure is recomputed from that run's per-drug output. Calcium-transient predictions come from the model's own cytosolic calcium state and have not been measured on cells.