The introduction of automated planar patch clamp instruments over the past two years has increased the throughput of voltage clamp ion channel assays by a factor of at least ten. This is possible because the automated systems can perform assays in parallel using16 and 384-well plates. While the drug discovery industry has embraced this new technology, the enthusiasm has been tempered by the modest success rates of the assays and by the high cost of the consumable patch substrate. Currently, typical success rates for a standard ion channel assay, using, for example, the Q-Patch from Sophion Biosciences, the Port-a-Patch system from Nanion Biosciences, or the PatchXpress from Molecular Devices Corp., is around 50%. In other words, for every16 channel chip used in these systems, only eight will produce useable data. This effectively doubles the price of each data point over what is ideally possible. In order for a planar patch clamp experiment to succeed, several events need to occur (assuming that the cell expresses the appropriate ion channels in functional states): the cell of interest must form a high-resistance seal with the planar substrate, the whole-cell configuration must be achieved, and fluidic pathways must be intact so that compounds of interest maybe applied to the cell. A failure of anyone of these steps will result in no data collected from that well. We propose to optimize the first two steps in this process, namely, seal formation and entry into whole-cell recording configuration. We will use machine learning approaches to examine how a human patch clamp expert interacts with the patch clamp system in order to develop a model that will provide parameters that can be used to more efficiently and successfully provide useable whole-cell recording configuration. It is important to note that the model that we derive from our approach will not actually copy what the expert does, but will attempt to optimize the process based on cues that mayor may not be consciously monitored by the expert. The Specific Aims of the Phase I component will be to: (1) integrate recording capabilities into existing automated patch clamp software from Nanion, (2) evaluate the success rate of the procedure specified by our machine learning analysis, and (3) develop stand-alone software for use specifically with manual patch clamp setups and for exploration of the potential benefits of using machine learning via expert training in other applications. In Phase II we propose to develop the proof-of-concept software into a user-friendly commercial software module which we will offer to existing and potential automated patch clamp companies. We will also simplify and streamline the user interface of this software as a stand-alone component for manual patch clamp systems. Developing drugs that target ion channels has been hindered by the expense of the consumables used in automated patch clamp screening devices. We propose to develop a method, using machine learning techniques which may increase the success rate of these instruments and therefore lower the overall cost of ion channel drug discovery
