Peak Detection#
Peak detection is one of the core features of snazzy_analysis.
From the detected peaks, we derive most of the metrics used in this package: peak widths, amplitudes, rise times, decay times, and more.
The algorithm consists of several steps, each with parameters that can be fine-tuned for optimal detection.
Each step is implemented as a single function that takes a params dictionary and changes the data in Trace._peak_idxes.
If you want to change or extend the peak detection steps, just add another function that follows this interface as a new stage inside Trace.detect_peaks.
To understand how the different parameters influence the peak detection, it’s important to first understand the entire peak detection algorithm.
1. Peak Detection on Low-Passed Filtered Signal#
The ΔF/F (DFF) trace is filtered in the frequency domain using a freq_cutoff parameter: all frequencies above this value are removed, and the remaining low-frequency components are used to reconstruct the filtered trace.
This acts as a smoothing step, which almost completely removes transients that do not correspond to the main activity bursts.
The freq_cutoff can be adjusted in the GUI using a slider, and the reconstructed signal is updated in real time.
The default value of 0.0025 Hz works well for many traces, but traces with high-frequency noise may require a lower cutoff.
Different types of samples will likely result in different traces and this value will have to be adjusted.
Once we have the filtered trace, peaks are detected using the parameters fft_height and fft_prominence.
The fft_height parameter is especially important because the reconstructed signal often contains minor ripples before the first real burst.
These are easy to identify, as they usually do not correspond to peaks in the original ΔF/F trace.
fft_prominence complements fft_height by measuring how much a point must stand tall from its surrounding baseline in order to be marked as a peak.
2. Align Peaks in the Original Signal#
After detecting peaks in the filtered signal, the peak indices are mapped back to the original ΔF/F trace. This step is necessary because the low-passed filter will result shift peak positions. The bursts of activity have a sharp rise and are followed closely by shorter transients. To properly mark bursts, we use the leftmost peak in each burst as the peak index.
The window size used to search for the leftmost peak is given by port_peaks_window_size.
Since the leftmost peak can have an amplitude very different than the local maximum peak, we specify the percentage from the local maximum we accept using the parameter port_peaks_thres.
3. Filter peaks by local threshold#
This step is not enabled by default.
To enable it the dff_peak_indices in trace.calculate_peaks must be passed to trace.filter_peaks_by_local_threshold.
As the embryos develop, there is a global trend of peak amplitude to rise and then stabilize before hatching. We use this fact to perform an extra validation step for the calculated peaks. Each peak is compared against its neighboring peaks, and peaks that are too high or too low are discarted. For example, if a peak close to baseline level is misidentified between bursts, it will likely be discarded due to all other peaks having higher values nearby.
The window size used to compare each peak with its neighbors is controlled using local_thres_window_size.
The minimum value for accepting a peak is given by local_thres_value.
4. Optional Post-Processing#
Some post-processing operations can further improve peak detection for specific types of traces.
Certain traces may exhibit abnormal transients at the beginning of the imaging session.
This is an artifact that should be removed.
In such cases, the remove_transients function can be applied.
It detects and removes initial bursts if their interval is significantly longer than the average of subsequent bursts.
Another post-processing step removes low-amplitude peaks that are likely false positives. Peaks below a specified percentage of the maximum peak amplitude are discarded.
Finally, in the GUI, you can manually add or remove peaks. These manual edits are used to update the set of calculated peaks.