We agree with the comment in Kleiman, Shah, and Morganroth (2014), that “[computer models]… need to be standardized, regulated and widely available before they are adopted to support sponsor and regulatory decisions”. It is sensible to ask “which

ion channels should we screen”? We consider important factors in the answer to this in the sections below. For our output of interest, how much can block of a particular channel influence the predictions? In this case, we are interested in predicting APD changes, it is evident from Fig. 2 that (depending on the model choice) IKr, ICaL and perhaps IKs block could have large effects on APD. At the degree of block likely to be encountered, block of (solely) INa and Ito have much less impact than those of the other channels, and so a choice could be made not to screen these. But more mechanistic predictions of pro-arrhythmic risk, Adriamycin other than simply APD prolongation, may be sensitive to the apparently-small changes we observed. Indeed, sodium channel blockers have been seen to prolong the QRS complex, potentially leading to increased pro-arrhythmic risk via conduction slowing or block, rather than delayed repolarisation (Gintant, Gallacher, & Pugsley, 2011). It is also worth noting that APD is not a linear function of channel block — blockade of INa and Ito could have large effects when another channel

is also being blocked. A more ‘global’ evaluation of the simulation output’s sensitivity to each channel block (under the influence of different combinations of block on the other channels) would be needed before concluding a channel cannot significantly see more found influence the outcome of interest. In contrast, additional ion channels — such as IK1 — can have a large effect on the action potential (Fig. 2). But these channels may not be blocked by a large enough proportion of compounds to consider screening them as standard, as discussed below. Some ion channels, pumps and exchangers are historically blocked by very few compounds. The outcome of ‘missing an effect’ in these rare cases is likely to be no more severe than progressing such a compound to later,

more expensive, safety testing, and picking up the effect there. The economic cost of screening for additional effects on such ion currents may therefore outweigh the cost of missing an ion current effect. There is also the variability, sensitivity and specificity of such screens to consider. In the case of an ion channel being blocked by as few as 1 in 10,000 compounds, the chance of a positive screening result being a ‘false positive’ is likely to far outweigh the chance of it being a ‘true positive’. A cost benefit analysis could be performed for each ion channel screening assay, based on: its variability, sensitivity and specificity; historical compound liability; and the cost of ‘missing’ an adverse interaction with this channel, and progressing to the next stage of testing.