A neural network model enables worm tracking in challenging conditions and increases signal-to-noise ratio in phenotypic screens
Fig 4
DTC improves the signal to noise ratio in a phenotypic screen.
All data from ref. [26]. (A) Speed calculated from Tierpsy skeletons and from DTC skeletons for a random sample of 290 wells from a previously published drug screen. The dashed line is y = x. (B) Correlation coefficients for each of the Tierpsy 256 behavioural features for the data from ref. [26] (left). The red line indicates the modal value of 0.71. Correlation coefficient plotted against the F-statistic for each feature calculated over the entire dataset. (C) Tail curvature as a function of dose for worms treated with a spiroindoline known to cause coiling. (D) A comparison of the mean absolute value of Hedge’s d effect size calculated using features derived from Tierpsy and DTC tracking data. Each point is the mean Hedge’s d across all doses of a drug compared to DMSO controls for a single feature. Any feature with ‘curvature’ in the name except for time derivatives of curvature are shown in blue. All other features are shown in yellow.