Import Dependencies¶

Download and Open NWB file¶

The NWB files can be found in Dandiset 001424 on DANDI. The NWB can be downloaded with the DANDI client and then opened with pynwb

The NWB and its sections are rendered in the notebook here for easy exploration. The most important sections are intervals which contains the stimulus information, and processing which contains the processed ophys recording.

Viewing the Traces¶

Information in exptSummary.E{}¶

The ophys module contains traces for the following fields:

• dF ('delta F') : Baseline-subtracted fluorescence
• F0: An estimate of the fluorescence baseline
• dFF ('delta F over F') : Baseline-subtracted fluorescence divided by baseline, i.e. fractional change in fluorescence.

For SLAP2, there are two simultaneously-imaged fields of view, and the traces in the each should have the same or very similar numbers of timepoints. For analysis, the two FOVs can be merged- the data are interpolated onto the same timebase.

Tip:
dF vs dFF When performing analyses of glutamate imaging, the more useful variable is dF, which has units proportional to photons.

For example, a good estimate of the total measured glutamate input to the cell is the sum of dF across synapses. It is inappropriate to sum dFF in the same way, because very dim synapses can have large dFF values and background fluorescence (e.g. from the dendritic shaft) strongly affects fractional changes at some synapses compared to others. This is different than population Calcium imaging, which tends to work with dF/F.

In glutamate imaging, dFF is most useful for compensating for bleaching over time for a single synapse, for example to see if its activity increases vs decreases over the course of an experiment. However even for this some confounds remain due to brightness-dependent bias in F0 estimation.

Reproducing F0 and DFF¶

The f0 and dff in these files were calculated as below. The code was translated from matlab using chatGPT from this file, written by Kaspar Podgorski.

The algo1 function is designed to denoise and extract a smooth baseline signal (F₀) from a noisy fluorescence trace. It first applies a median filter to suppress noise and transient spikes, with the window size defined by denoiseWindow. Next, it uses a rolling convex-hull-like operation: the signal is sampled at intervals, and at each sample, multiple interpolated estimates are generated using subsampled windows, from which the minimum is taken to form a robust lower envelope. To further ensure smoothness and stability, unreliable points—especially near gaps or NaNs—are identified and replaced using local moving averages. The result is a smooth, denoised baseline that can serve as a reference for downstream analysis, such as dF/F computation.

To compute dF/F (delta F over F), the baseline fluorescence $F_0$ is subtracted from the raw signal $F$, and the result is divided by $F_0$:

$$ \text{dF/F} = \frac{F - F_0}{F_0} $$

This normalizes the signal to highlight relative changes in fluorescence.