Comments from curator
Phenotyping data comes from the authors' Zenodo archive (
10.5281/zenodo.13698148, also linked from the
lab page; analysis code on
GitHub), file
Drift Data/DGRP Data/DGRPsummerall.csv of Drift-in-Individual-Preference.zip. It has one row per fly (batch, tray, fly number, DGRP line, sex code) with, for each of the 9 test days, the number of turns (n) and the fraction of right turns (rBias). The authors' notebook DGRPStan.ipynb fits its Bayesian model to this file.
Design: flies of 10 DGRP lines (Table 1 of the
paper: DGRP 45, 85, 105, 208, 426, 535, 703, 796, 819 and 907, with their Bloomington numbers) were reared on cornmeal/dextrose food (12:12 h light:dark, 20-23 °C) and kept alone (vials or individual housing plates) so that each fly could be followed. Each fly was tested in a Y-maze (MARGO tracking, white LED light, 20 min acclimation after CO
2 anaesthesia) for 2 h on 3 days a week for 3 weeks, starting 3-6 days after eclosion. Its handedness on a day is the fraction of right turns. The paper uses these data to estimate, per line, the initial variability of handedness among flies (bet-hedging), its day-to-day change (drift) and its persistence (φ), with a hierarchical Bayesian autoregressive model (Fig. 1E-H).
Raw dataset (raw_1), one row per fly (1,193 flies, 22 to 235 per line):
-
NumTurns_WkDj and
TurnBias_WkDj: number of turns and fraction of right turns on test day j (1-3) of week k (1-3), copied from the file. In the file, rBias times (n - 1) is a whole number of right turns for every record. A count of -1 (1,523 records: fly not measured that day; 144 flies have values again later) and empty days (3,360 records: whole trays not tested on that day) are NA. Counts of 0 (224 records) are kept; TurnBias is NA for flies with 0 or 1 turn. No minimum number of turns is applied: the authors kept flies with more than 50 turns in Fig. 1E-F, and their model uses all turns (binomial). The number of flies tested drops over the 3 weeks, and some lines were not tested on some days (DGRP_085 after W2D1, DGRP_703 on W1D3 and in week 3, DGRP_907 on W2D3).
-
Batch,
Tray,
FlyID (obsolete design columns): the batch (1-5), tray of 96 flies and fly number of the file.
-
SexCode (obsolete): the file codes sex as 0 or 1 (1,059 and 134 flies) but neither the file, the paper nor the code gives the key, so
sex is NA for every fly. The authors also pooled the sexes. Lines 703 and 907 have only code 0.
Not uploaded: the 55 flies of 'LineNumber 0' (not a DGRP line, not in Table 1, removed by the authors' notebook). The per-line model estimates (σ
Bet-hedging, σ
Drift, φ; Fig. 1H) are only shown as plots. The continuous circling (Canton-S), serotonin (Isod1,
trh mutants) and simulation parts of the paper do not use DGRP lines.
Mismatches with the paper: Table 1 gives 55 flies for DGRP 45, but the file and the notebook have 145 (the other 9 lines match Table 1 exactly). The Fig. 1E legend says n = 48-235 per genotype, but DGRP 85 has 22 flies. The archived notebook melts only the first 23 columns, which leaves out rBias of the last test day (W3D3), so that day is treated as missing in its model fit (5,340 observations, reproduced; 5,558 with W3D3). Its other outputs are reproduced from the uploaded file (1,193 flies; 882 flies with more than one day of > 50 turns; per-fly mean and SD of right bias). The Y-maze data of study 203 (de Bivort et al. 2022, same lab) are different flies.