9. Analyze: feeding analysis

The Analyze page turns recordings into feeding measures, statistics, figures and Excel tables. One analysis can combine several recordings. It analyses every CapacitanceData file in the chosen folder, so a folder holds one complete experiment.

Start an analysis

• From Review: Analyze this recording → (recommended; the folder and labels are pre-filled).
• On the Analyze page: New Analysis, or Analysis → New analysis… (Ctrl+N).
• To reopen a finished analysis: Load Existing or Analysis → Load analysis results… (Ctrl+L).

The setup wizard

1. Data Location. The folder with your CapacitanceData files; the wizard reports how many it found. Optionally, choose an output file to save the analysed events for later.

2. Hardware Configuration.

3. Experiment Labels. Food (substrate) and condition names. They are filled in from the Log_…txt file saved with the recording; Load from Log File… reads another one.

4. Analysis Parameters. The defaults suit most experiments:

Sip detection. The same page offers two versions (also in Analysis → Sip detection; the choice is remembered):

• V2 – current version (recommended): a sip’s touch and release steps must be of similar size, and the activity threshold adapts to the noise of each channel. This removes onsets placed about 100 ms early on detrending echoes, and spurious sips.
• V1 – MATLAB: identical to the MATLAB analysis, for comparing with results analysed there.

5. Review Settings. A summary. Click Finish to run.

Data checks

When processing finishes, a Data checks window lists anything unusual. The same list is saved as AnalysisWarnings.txt with the results.

When the conditions have their own food pairs (see Acquire), each plot and table shows each condition’s foods. Plots of one side are titled Food 1 (left) / Food 2 (right), and the preference index is labelled with each condition’s pair, e.g. C01 (1 mM sucrose vs 2 mM sucrose).

Long recordings (days)

Recordings of several days are analysed like any other. Set Duration to the recording length in samples: 360,000 per hour, e.g. 25,920,000 for 3 days (the maximum is 11.6 days).

• Memory. The file is never loaded whole. It is read in parts, a group of channels at a time, and each channel is analysed over its full length, so the results are the same as for a short file. A long recording is processed by several channels in parallel, as many as the processor cores and 60 % of the free memory allow (3–4 GB per channel for 3 days). A PC with little free memory simply uses fewer and takes longer; it never runs out of memory.
• Time. A 3-day, 96-channel recording on an 8-core PC with 32 GB: about 4 minutes of processing and 3½ minutes to write the figures, Excel and CSV files.
• Time-resolved per-fly workbooks (10-s bins): Excel holds at most 16,384 columns (45.5 h). For longer recordings the same grids are written as CSV files in the folder TimeResolvedCSV (one file per sheet, one row per fly, one column per 10-s bin, full precision), readable by MATLAB, Python, R or Prism. The analysis log says so.

Results

Processing takes seconds to a few minutes. Then the results viewer opens: plot categories on the left, the selected plot on the right.

Above each plot:

• Export saves the figure.
• Statistics shows the tests between conditions.
• Customize changes colours, labels and fonts.
• Show Data shows the numbers behind the plot.

Stimulation vs catch

When a recording was made with an OptoPad Probability below 1 (see OptoPad), the analysis reads its …_events.csv and compares, per fly, the moments that were stimulated with the matched moments that were not (catch trials). Each fly is compared with itself (paired Wilcoxon test, flies with at least one stim and one catch trial); the p value and n are printed on the figures. These figures are not in the Statistics window, which compares conditions.

Stim vs catch: trials — aligned to each trial moment, on the channel that triggered:

• Peri-trial sip rate: sips per second from −10 to +30 s, stim and catch curves (mean ± SEM over flies) per condition.
• Trial measures: sips in the 10 s after the trial; time until feeding stops (end of the current activity bout, 0 if the fly was not feeding); latency to feed again (from the end of that bout to the next one). Each fly’s value is the mean over its trials.

The windows are set in the wizard (Stim vs catch rows).

Stim vs catch: periods — everything the fly did during stim periods vs during catch periods, stim and catch side by side per condition, each fly’s two values joined:

• every box-plot and microstructure measure, per food;
• the preference index (sips and activity) and transition IBIs;
• sip-duration and inter-sip histograms.

How the periods are counted:

• With two foods, both channels of a fly share its periods (the LED lights the arena). This holds even when you remove one of the two channels.
• USB dropouts are cut out of the periods: they count neither events nor time.
• An event belongs to the period in which it starts. An interval (between sips, bouts or bursts) counts only when both ends lie in the same period.
• Counts and totals are per minute of period (a fly may have more catch time than stim time); medians are unchanged.
• The recording-long cumulative time courses are not repeated for stim vs catch: periods are short and scattered, so such a curve would mostly show when trials happened. The peri-trial sip rate is the time course aligned to the trial instead.
• A recording with stimulations but no catch rows (probability 1, or recorded before catch trials were logged) gets a Data checks warning and no stim vs catch figures.

All numbers are saved in StimVsCatchInExcelFormat.xlsx: one row per trial (Trials), per-fly means (PerFly), minutes and number of periods per fly (Periods), the peri-trial curves (PeriTrialRate), every period measure per fly (PeriodMetrics) and the tests (Stats).

Saved outputs

Every analysis automatically saves all figures and tables to figures and datasheets in the data folder:

Extra per-fly outputs (identical to the MATLAB “extra outputs” analysis):

Rows marked IsPhantomRow = 1 in FlyIndexInExcelFormat.xlsx are padding, not flies. Drop them before statistics.

Statistics and the preference index

PI = (sips on A − sips on B) / (sips on A + sips on B) ranges from +1 (only A) to −1 (only B). The analysis adds 1 to each count, (A − B) / (A + B + 2), so that flies that barely fed sit near 0 instead of at ±1.

The Statistics window compares conditions with one value per fly: the Mann–Whitney U test between each pair of conditions (optionally Bonferroni or Holm corrected), the Kruskal–Wallis test or one-way ANOVA across all conditions, and Cliff’s delta and Cohen’s d as effect sizes.