Average of Baseline Autocorrelation Function is a Leading Indicator of Neural Stimulation Data Quality

The quality of data recorded across experimental neural stimulation sessions can vary significantly, and in some cases data from lower quality sessions costs us time, effort, and statistical power. That fact creates the need for leading indicators of our data quality - indicators which can be measured very early and cheaply in the session, which can potentially be acted on to improve the data quality. Here we present such a leading indicator which we found to be predictive of the quality of data measuring the neural response to stimulation across 40 excitatory optogenetic stimulation sessions. Our indicator is based on the average of the autocorrelation function over multiple lags, which we calculate on ≈5min of baseline (i.e. non-stimulation) data at the beginning of the session. The indicator measures the degree of correlation in a signal, and may be a proxy to the signal's noise level. We show that our indicator is highly correlated with our ability to forecast the neural response to stimulation later in the session, even using complex non-linear dynamics models based on long short-term memory (LSTM) networks, and temporal basis function models (TBFMs). As a result, it provides a potentially actionable leading indicator which is fast and cheap to acquire.Clinical relevance- By providing a leading indicator, our method may give us actionable intelligence early in experimental sessions. That may allow us to detect opportunities to improve data quality, potentially preventing the waste of valuable time and effort.