User Tools

Site Tools


Manual validation of 1x freely moving pumping experiments

Nicolina counted 3 animals as they approach a lawn (1 worm was not tracked in automatic, and is not used. Could be added later). These movies were also analyzed with pharaglow. We selected only animals that successfully entered during the recording. Timestamps between the manual counts and the automated tracking were aligned and a few metrics were calculated. Overall, we can find a common pre-processing and peak detection parameter set that performs well for all 3 animals. The deviations were

Auto Manual Deviation
worm 2 158 161 3
worm 3 160 155 -5
worm 4 149 147 -2


Worm 2

Worm 3

Worm 4

(A) Cumulative pumping rate for the automatic and manually tracked data (blue and orange, respectively)
(B) Pumping rate for both cases (calculated as instantanous rate) and a smoothed version (1 sec smoothing). The dashed line denotes lawn entry.
(C) Correlation between automated and manual pumping rate. The black line shows unity, and the red and blue line show the expected error ranges for different frequencies based on a the manual annotator deviating its pumps by a single frame. This effect becomes pronounced at large pumping rates. e.g., if there are true pumps 5 frames apart, this would lead to a rate of 30fps/5frames = 6Hz. However, if the manual annotator deviated by 1 frame to result in the peaks being 6 frames apart, the frquency would be 30 fps/6 = 5Hz.
(D) Histogram of pumping rates.
(E) velocity as the worm enters the lawn. The dashed line denotes lawn entry.
(F) Instantaneous deviations in pumps per 0.5s bin (black) and cumulative deviation of detected pumps from manual pumps.


  • The current tracking performs reasonably well, especially considering that we will average across many worms. Trends and smoothed pumping rates (1 sec smoothing) were looking good.
  • For analysis of the time series in more detail, we need to track individual animals. This should be done using the rtacking stage planned for calcium imaging anyway.
  • The automated analysis performs worse in the lawn. We need to check if this is due to a small focus issue and if yes, this might resolve itself with the larger depth-of-field in the cheap behavior microscopes.
  • Overall, we can move on with screens etc, based on the accuracy presented here
wiki/documentation/monika/manual_validation_of_1x_freely_moving_pumping_experiments.txt · Last modified: 2020/03/11 00:45 by mscholz