The nervous system integrates stimuli, internal states, expectations, and previous experience to regulate behavioural output. Describing, quantifying, and modulating behaviour are critical aspects of modern neuroscience and, ever since its inception, the field has spent considerable effort into building and sharing paradigms or tools aimed at objectively and reproducibly quantify behaviours in the most disparate animal models, to the point that this exercise is now recognised as an exciting subfield of neuroscience in its own right: ethomics (Brown and de Bivort, 2018; Datta et al., 2019). As the portmanteau name itself suggests, ethomics is not just about describing behaviour (etho-) but also about doing so in a high-throughput fashion (-omics), collecting data simultaneously from a large number of individuals, which can remain undisturbed throughout recording. Irrespective of the behaviour or the animal model to be analysed, the first compromise a researcher will face when choosing a tool for behavioural quantification will always be between throughput and resolution: a high-throughput analysis will allow for powerful experimental manipulations – such as genetics or pharmacological screens – offering unbiased approaches in identifying neuronal circuits, genes, molecules underpinning behaviour; high-resolution analysis, on the other hand, promises to identify and discriminate even minuscule differences that may not be immediately visible to the human eye, and to label behaviours into identifiable classes (e.g. ‘grooming’, ‘courting’, and ‘shaking’) that may be more relevant to researchers interested in modelling disease or in anthropomorphic descriptions. In the past years, the field has generally converged towards the adoption of high-resolution video recording of activity, in some cases adopting cameras that have milliseconds temporal resolution or developing setups that provide depth information for three-dimensional reconstruction of motion or posture (Wiltschko et al., 2015; Hsu and Yttri, 2021; Nath et al., 2019; Pereira et al., 2019; Gosztolai et al., 2021). Given the recent evolution in machine learning and progresses in computational power, even these high-resolution analyses can be at least in part compatible with high-throughput approaches (Wiltschko et al., 2020), especially when employed on small invertebrate animal models (McDermott-Rouse et al., 2021; Ayroles et al., 2015; Branson et al., 2009; Kabra et al., 2013) or when aided by robotic handling (Alisch et al., 2018). These systems, however, can still be prohibitively expensive for most laboratories, and not easily compatible with throughput in the ‘omics scale. Moreover, besides the technical urge of removing entry barrier and make ethomics an accessible tool, an equally important underlying question concerns what is the minimal amount of information that needs to be extracted to identify and classify behaviour. Do we always necessarily gain information from extracting micro-postural features or by analysing activity in three dimensions? To what extent this may actually add counterproductive biological noise to some assays?

Here, we introduce coccinella: a new experimental framework that combines high-throughput, inexpensive, real-time ethomics (Geissmann et al., 2017) with state-of-the-art statistical analysis (Fulcher and Jones, 2017; Lubba et al., 2019) to characterise and discriminate complex behaviours using a reductionist approach based solely on one simple feature (https://lab.gilest.ro/coccinella). Coccinella builds on ethoscopes (Geissmann et al., 2017), an accessible open-source platform, to extract, in real-time, activity information from flies. Despite its minimalist nature, coccinella outperforms state-of-the-art alternatives in recognising the pharmacobehavioural space, providing better discernibility at a fraction of the cost, thus opening a new path to high-throughput ethomics.

Drosophila ethomics studies generally rely on image acquisition through so-called industrial cameras, able to collect videos with high temporal and spatial resolution and featuring mounts for a large selection of lenses. Some of the cameras commonly used for these purposes (e.g. FLIR, Point Grey, Basler) (Mathis et al., 2018) are expensive and normally employed in close-up imaging that microscopically highlights the smallest anatomical features of the animal but at the same time greatly limits the number of experimental subjects that can be recorded by a single device. Normally, one or few more camera would be connected to a dedicated powerful computer for acquisition and storage of videos. The cost and physical footprint of these setups makes them incompatible, at least for most laboratories, with high-throughput simultaneous acquisition. To lower this barrier, we created a framework that employs the distributing computing power of ethoscopes (Geissmann et al., 2017), thus allowing for inexpensive analysis of activity in hundreds or thousands of flies at once. Ethoscopes are open source and can be manufactured by a skilled end-user at a cost of about £75 per machine, mostly building on two off-the-shelf component: a Raspberry Pi microcomputer and a Raspberry Pi NoIR camera overlooking a bespoke 3D-printed arena hosting freely moving flies. The temporal and spatial resolution of the collected images depends on the working modality the user chooses. When operating in offline mode, ethoscopes are capable to acquire 720p videos at 60 fps, which is a convenient option with fast moving animals. In this study, we instead opted for the default ethoscope working settings, providing online tracking and real-time parametric extraction, meaning that images are analysed by each Raspberry Pi at the very moment they are acquired (Figure 1b). This latter modality limits the temporal resolution of information being processed (one frame every 444 ± 127 ms, equivalent to 2.2 fps on a Raspberry Pi3 at a resolution of 1280 × 960 pixels with each animal being constricted in an ellipse measuring 25.8 ± 1.4 × 9.85 ± 1.4 pixels – Figure 1a) but provides the most affordable and high-throughput solution, dispensing the researcher from organising video storage or asynchronous video processing for tracking the animals. In the work here described, flies moved freely in a circular two-dimensional space with a diameter of 11.5 mm designed to maintain the animal in the walking position (Simon and Dickinson, 2010) while venturing on solidified agar providing nutrients alone or nutrients and drugs. In previous analysis of activity and sleep (Geissmann et al., 2017; Geissmann, 2018), we found that the maximal velocity of the fly over a period of 10 s best described the basic motion features of the animal, allowing us to accurately differentiate between different activity patterns, such as walking, grooming, and feeding (Geissmann et al., 2017). We therefore adopted this measure for coccinella too, ultimately producing monodimensional time series of a behavioural correlate, which were then digested using highly comparative time-series analysis (HCTSA) (Fulcher and Jones, 2017), a computational framework that effortlessly subjects time series to more than 7700 literature-relevant statistical tests, looking for meaningful discriminative features. Features successfully extracted through HCTSA were then used to classify behaviour using a linear support vector machine (SVM_linear) (Chatterjee et al., 2022; Figure 1b) and here presented and compared using confusion matrices (Figure 1—figure supplement 1a). To test the ability of this system to discriminate behaviour, we started by exploring the pharmacobehavioural space of flies fed with a panel of known or putative neurotropic chemicals, comprising molecules previously described in the literature along with uncharacterised ones being considered as potential insecticides (Figure 1c, d and Supplementary file 1). Using an initial panel of 17 treatments (16 drugs and 1 solvent control) we were able to discern compounds with an accuracy of 71.4% (vs. 5.8% of a random classifier – Figure 1c). Some compounds induced behaviours with a particularly high predictive fingerprint: dieldrin, for instance, was predicted with an accuracy of 94% and flonicamid with an accuracy of 87%. For others, our system fared more poorly (e.g. tetramethrin showed 41% accuracy). In all cases, however, the relative confusion was negligible, with all compounds being correctly identified as the first choice and with the first predicted compounds having, on average, a score that was 15 times greater compared to the second-best choices (Figure 1c – min.: 3.3×; max.: 65×). To validate our framework and exclude artefacts operated by overfitting biologically irrelevant information we followed two lines of control. Firstly, we fed flies with lower concentrations of the same compounds (Figure 1c). Feeding flies with different concentrations of drugs unsurprisingly showed a different effect on short-term lethality (Figure 1—figure supplement 1b), with several compounds hitting a 25% lethality rate before the end of the experiment when fed at the highest concentration (1000 ppm – Figure 1—figure supplement 1b). Lowering the compound concentration, the predictive accuracy of the system decreased from 71.4% (1000 ppm) to 61.8% (100 ppm), falling to 36.1% with the lowest concentrations (1 ppm), indicating that the system does operate on pharmacologically induced, biologically meaningful behavioural correlates (Figure 1c). A similar drop in accuracy was observed using a smaller panel of 12 treatments (Figure 1—figure supplement 1c). As a second line of work to test specificity, we obtained genetic mutants known to be resistant to specific pharmacological treatments: the para^L1029F allele encodes for a version of the α-subunit of voltage-gated sodium channel conferring resistance to dichlorodiphenyltrichloroethane (DDT) and pyrethroids (Kaduskar et al., 2022); the Rdl^A301S allele encodes for a version of the ligand-gated chloride channel conferring resistance to dieldrin and fiproles (Remnant et al., 2014). Challenging these mutants with their respective compounds created confusion in the clustering algorithm for which discerning between drug treatment and solvent control became a harder task, especially in the case of DDT and para^L1029F (Figure 1—figure supplement 2a). The observed drop in accuracy suggests again that coccinella is working on biologically relevant behavioural signatures and the fact that some discrimination can still be observed with targets harbouring point mutations – which should severely affect compound efficacy – is indicative of high sensitivity.

Figure 1 with 2 supplements see all

Download asset Open asset

Having established the accuracy and sensitivity of the system, we next wanted to test its usefulness in a genuine high-throughput scenario. We subjected a total of 2192 flies to a panel of 40 treatments (Figure 1d), mostly featuring known compounds but also two unexplored molecules (Supplementary file 1). Given that the 100 ppm intermediate concentrations showed the best compromise between accuracy and lethality in the previous pilot experiment (Figure 1—figure supplement 1b), we performed this larger screen using compounds diluted at 100 ppm only. Even with such a large panel, the system was able to first-guess 39 out of 40 of the tested treatments (the only exception being the Syngenta Compound #5) with an overall accuracy of 44.5% vs. 2.5% of the random classifier.

A reductionist approach served us well so far, identifying with remarkable accuracy even subtle changes when we explored the pharmacobehavioural space of a large number of neuroactive compounds in wild-type and mutant flies. But how does it compare to other more established paradigms? Coccinella is arguably to be preferred in terms of accessibility and throughput, but what is the amount of useful information that we are sacrificing by adopting a reductionist approach? To quantify any possible loss in information content, we ran a series of parallel experiments in which flies were fed with a selected panel of 12 treatments (11 drugs and a solvent control, Figure 2) and their behaviour analysed either using coccinella or using other widely adopted state-of-the-art methods, which started with high-resolution imaging and employed supervised machine learning for pose-estimation DeepLabCut; Mathis et al., 2018 followed by unsupervised identification of behavioural grammar (B-SOiD; Hsu and Yttri, 2021). To widen the range of comparisons, data were then either immediately clustered using different common clustering algorithms (K-nearest neighbours, random forest) or first processed through a smaller, selected subsample of the HCTSA array (Catch22; Lubba et al., 2019) before being clustered using SVM_linear (Figure 2a). In this challenge, coccinella unambiguously identified 10 out of 12 compounds, with poor performance only for two of them (flubendiamide and tetramethrin) and an overall accuracy of 42.4% vs. 8.3% of the random classifier (Figure 2b). Surprisingly, none of the state-of-the-art high-resolution paths did better than this. The combination of pose-estimation → grammar extraction → random forest classification scored as the second best, with an accuracy of 25.4% but with only three compounds being unambiguously identified (Figure 2c). The same experimental dataset clustered with even poorer performance when using K-nearest neighbours (Figure 2e) and even the application of HCTSA features extraction to the B-SoID output still could not match the accuracy observed with coccinella’s reductionist approach (Figure 2d). This analysis is not meant to be conclusive. We expect that some alternative combination of state-of-the-art approaches will probably manage to match or likely improve over coccinella’s performance, yet the fact we could obtain such an impressive result with a system that is arguably unmatched in terms of throughput and economic cost is, alone, an argument that gives new weight to this (and future) reductionist approaches.

Figure 2

Download asset Open asset

Finally, to push the system to its limit, we asked coccinella to find qualitative differences not in pharmacologically induced changes in activity, but in a type of spontaneous behaviour mostly characterised by lack of movement: sleep. In particular, we wondered whether coccinella could provide biological insights comparing conditions of sleep rebound observed after different regimes of sleep deprivation. Drosophila melanogaster is known to show a strong, conserved homeostatic regulation of sleep that forces flies to recover at least in part lost sleep, for instance after a night of forceful sleep deprivation (Shaw et al., 2000; Hendricks et al., 2000). We previously showed that the extent of sleep rebound observed after sleep deprivation only loosely correlates with the amount of lost sleep (Geissmann et al., 2019a) and it remains an open question whether similar amounts of sleep rebound may in fact differ from each other in some inscrutable feature that would underpin a different ‘sleep depth’ (Wiggin et al., 2020; French et al., 2021), similar to what it is believed to happen in mammals. Here, we analysed a dataset of 727 flies that experienced different regimes of mechanically enforced sleep deprivation during the 12 hr of the night (Figure 3). Flies were housed in tubes that would rotate after a set time of inactivity ranging from 20 to 1000 s leading to different degrees of sleep restriction (Figure 3a, dataset from Geissmann et al., 2019a). In this experimental paradigm, all treatments led to a statistically significant rebound compared to the undisturbed control animals (Figure 3b). We then ran coccinella on the two subsets of the panel: the baseline data, acquired the morning before the sleep deprivation (Figure 3c), and the rebound data, on the morning after (Figure 3d). Unsurprisingly, we could not detect any internal biological difference in the pre sleep deprivation control set, featuring flies of identical genotype and age housed in different tubes before the sleep deprivation treatment. In these conditions, coccinella could not discern, and performed exactly as a random classifier would (9% vs. 9% – Figure 3c). However, analysis of those same animals during rebound after sleep deprivation showed a clear clustering, segregating the samples in two subsets with separation around the 300 s inactivity trigger (Figure 3d). This result is important for two reasons: on one hand, it provides, for the third time, strong evidence that the system is not simply overfitting data of nought biological significance, given that it could not perform any better than a random classifier on the baseline control. On the other hand, coccinella could find biologically relevant differences on rebound data after different regimes of sleep deprivation. Interestingly enough, the 300 s threshold that coccinella independently identified has a deep intrinsic significance for the field, for it is considered to be the threshold beyond which flies lose arousal response to external stimuli, defining a ‘sleep quantum’ (i.e. the minimum amount of time required for transforming inactivity bouts into sleep bouts; Shaw et al., 2000; Hendricks et al., 2000; Joyce et al., 2023). Coccinella’s analysis ran agnostic of the arbitrary 5 min threshold and yet identified the same value as the one able to segregate the two clusters, thus providing an independent confirmation of the 5-min rule in D. melanogaster.

Figure 3

Download asset Open asset

System neuroscience is living a period of renaissance, and Drosophila is driving this revolution strong of the first full-brain connectome, a plethora of new genetic reagents that allow thermo- and opto-genetic manipulations, a galore of genetic transformants for circuit tracking and manipulation, and multiple tools for large-scale quantification of behaviour. Progresses in machine learning and computer power have had a massive impact on the field of ethomics, especially in achieving levels of anatomical tracking that allow mapping of the tiniest movements on an experimental animal model with the highest temporal resolution and with little human supervision (Pereira et al., 2019; Mathis et al., 2018). Most of these systems, however, rely on relatively expensive setups and do not scale easily to high-throughput experimental paradigms. They are ideal – and irreplaceable – to identify behavioural patterns and study fine motor control but may be undue for other uses. Here, we introduce a new framework, coccinella, that merges an open-source, economically accessible hardware platform (ethoscopes; Geissmann et al., 2017; Geissmann et al., 2019b) with a powerful toolbox for statistical analysis and clustering (HCTSA, Fulcher and Jones, 2017/Catch22, Lubba et al., 2019). Coccinella is a reductionist tool, not meant to replace the behavioural categorisation that other tools can offer but to complement it. It relies on Raspberry PIs as main acquisition devices, with associated advantages and limitations. Ethoscopes are inexpensive and versatile but are limited in terms of computing power and acquisition rates. Their online acquisition speed is fast enough to successfully capture the motor activity of different species of Drosophilae (Joyce et al., 2023), but may not be sufficient for other animals moving more swiftly, such as zebrafish larvae. Moreover, coccinella cannot – and is not meant to – apply labels to behaviour (‘courting’, ‘lounging’, ‘sipping’, ‘jumping’, etc.) but it can successfully identify large behavioural phenotypes and generate unbiased hypothesis on how behaviour, and a nervous system at large, can be influenced by chemicals, genetics, artificial manipulations in general. Here, we provided evidence that coccinella can be used to successfully explore and compartmentalise the pharmacobehavioural space and also showed that a reductionist approach can be employed to discern otherwise invisible shades of a very subtle naturally occurring behaviour: sleep. The success of Drosophila as experimental model was built on the many genetic screens of the 1900s. We propose coccinella as an accessible, pivotal tool to boost again this important line of work in any laboratory, without funding or access to technology being a discriminative factor.

A notebook version of the source code used to generate all figures is available on the Zenodo public repository, along with all the metadata and the raw data collected in this study (DOIs: 10.5281/zenodo.7335575 and 10.5281/zenodo.7393689). Data were analysed using rethomics (Geissmann et al., 2019b) and ethoscopy (Blackhurst et al., 2023). Two notebooks showing how to use the system for multiple uses are provided in Supplementary file 2.

1. 1. Jones H
   2. Gilestro G

   (2022) Zenodo

   A reductionist paradigm for high-throughput behavioural fingerprinting in Drosophila melanogaster - DATASET 1 of 2.

   https://doi.org/10.5281/zenodo.7335575
2. 1. Jones H
   2. Gilestro G

   (2022) Zenodo

   A reductionist paradigm for high-throughput behavioural fingerprinting in Drosophila melanogaster - DATASET 2 of 2.

   https://doi.org/10.5281/zenodo.7393689

1. 1. Alisch T
   2. Crall JD
   3. Kao AB
   4. Zucker D
   5. de Bivort BL

   (2018) MAPLE (modular automated platform for large-scale experiments), a robot for integrated organism-handling and phenotyping

   eLife 7:e37166.

   https://doi.org/10.7554/eLife.37166

   • PubMed
   • Google Scholar
2. 1. Arena JP
   2. Liu KK
   3. Paress PS
   4. Schaeffer JM
   5. Cully DF

   (1992) Expression of a glutamate-activated chloride current in Xenopus oocytes injected with Caenorhabditis elegans RNA: evidence for modulation by avermectin

   Brain Research. Molecular Brain Research 15:339–348.

   https://doi.org/10.1016/0169-328x(92)90127-w

   • PubMed
   • Google Scholar
3. 1. Ayroles JF
   2. Buchanan SM
   3. O’Leary C
   4. Skutt-Kakaria K
   5. Grenier JK
   6. Clark AG
   7. Hartl DL
   8. de Bivort BL

   (2015) Behavioral idiosyncrasy reveals genetic control of phenotypic variability

   PNAS 112:6706–6711.

   https://doi.org/10.1073/pnas.1503830112

   • PubMed
   • Google Scholar
4. 1. Blackhurst L
   2. Gilestro GF
   3. Yu G

   (2023) Ethoscopy and ethoscope-lab: a framework for behavioural analysis to lower entrance barrier and aid reproducibility

   Bioinformatics Advances 3:vbad132.

   https://doi.org/10.1093/bioadv/vbad132

   • PubMed
   • Google Scholar
5. 1. Blythe J
   2. Earley FGP
   3. Piekarska-Hack K
   4. Firth L
   5. Bristow J
   6. Hirst EA
   7. Goodchild JA
   8. Hillesheim E
   9. Crossthwaite AJ

   (2022) The mode of action of isocycloseram: A novel isoxazoline insecticide

   Pesticide Biochemistry and Physiology 187:105217.

   https://doi.org/10.1016/j.pestbp.2022.105217

   • Google Scholar
6. 1. Branson K
   2. Robie AA
   3. Bender J
   4. Perona P
   5. Dickinson MH

   (2009) High-throughput ethomics in large groups of Drosophila

   Nature Methods 6:451–457.

   https://doi.org/10.1038/nmeth.1328

   • PubMed
   • Google Scholar
7. 1. Brown AEX
   2. de Bivort B

   (2018) Ethology as a physical science

   Nature Physics 14:653–657.

   https://doi.org/10.1038/s41567-018-0093-0

   • Google Scholar
8. Book

   1. Chatterjee S
   2. Dey D
   3. Munshi S

   (2022) Chapter 4 - feature selection and classification

   In: Chatterjee S, Dey D, Munshi S, editors. Recent Trends in Computer-Aided Diagnostic Systems for Skin Diseases. Academic Press. pp. 95–135.

   https://doi.org/10.1016/B978-0-323-91211-2.00001-9

   • Google Scholar
9. 1. Datta SR
   2. Anderson DJ
   3. Branson K
   4. Perona P
   5. Leifer A

   (2019) Computational neuroethology: A call to action

   Neuron 104:11–24.

   https://doi.org/10.1016/j.neuron.2019.09.038

   • PubMed
   • Google Scholar
10. 1. Dong K

    (2007) Insect sodium channels and insecticide resistance

    Invertebrate Neuroscience 7:17–30.

    https://doi.org/10.1007/s10158-006-0036-9

    • PubMed
    • Google Scholar
11. 1. Ffrench-Constant RH
    2. Roush RT
    3. Mortlock D
    4. Dively GP

    (1990) Isolation of dieldrin resistance from field populations of Drosophila melanogaster (Diptera: Drosophilidae)

    Journal of Economic Entomology 83:1733–1737.

    https://doi.org/10.1093/jee/83.5.1733

    • PubMed
    • Google Scholar
12. 1. French AS
    2. Geissmann Q
    3. Beckwith EJ
    4. Gilestro GF

    (2021) Sensory processing during sleep in Drosophila melanogaster

    Nature 598:479–482.

    https://doi.org/10.1038/s41586-021-03954-w

    • PubMed
    • Google Scholar
13. 1. Fulcher BD
    2. Jones NS

    (2017) hctsa: A computational framework for automated time-series phenotyping using massive feature extraction

    Cell Systems 5:527–531.

    https://doi.org/10.1016/j.cels.2017.10.001

    • PubMed
    • Google Scholar
14. 1. Geissmann Q
    2. Garcia Rodriguez L
    3. Beckwith EJ
    4. French AS
    5. Jamasb AR
    6. Gilestro GF

    (2017) Ethoscopes: An open platform for high-throughput ethomics

    PLOS Biology 15:e2003026.

    https://doi.org/10.1371/journal.pbio.2003026

    • PubMed
    • Google Scholar
15. Book

    1. Geissmann Q

    (2018) High-Throughput Recording, Analysis and Manipulation of Sleep in Drosophila

    Imperial College London Press.

    https://doi.org/10.25560/69514

    • Google Scholar
16. 1. Geissmann Q
    2. Beckwith EJ
    3. Gilestro GF

    (2019a) Most sleep does not serve a vital function: Evidence from Drosophila melanogaster

    Science Advances 5:eaau9253.

    https://doi.org/10.1126/sciadv.aau9253

    • PubMed
    • Google Scholar
17. 1. Geissmann Q
    2. Garcia Rodriguez L
    3. Beckwith EJ
    4. Gilestro GF

    (2019b) Rethomics: An R framework to analyse high-throughput behavioural data

    PLOS ONE 14:e0209331.

    https://doi.org/10.1371/journal.pone.0209331

    • PubMed
    • Google Scholar
18. 1. Gosztolai A
    2. Günel S
    3. Lobato-Ríos V
    4. Pietro Abrate M
    5. Morales D
    6. Rhodin H
    7. Fua P
    8. Ramdya P

    (2021) LiftPose3D, a deep learning-based approach for transforming two-dimensional to three-dimensional poses in laboratory animals

    Nature Methods 18:975–981.

    https://doi.org/10.1038/s41592-021-01226-z

    • PubMed
    • Google Scholar
19. 1. Hendricks JC
    2. Finn SM
    3. Panckeri KA
    4. Chavkin J
    5. Williams JA
    6. Sehgal A
    7. Pack AI

    (2000) Rest in Drosophila is a sleep-like state

    Neuron 25:129–138.

    https://doi.org/10.1016/s0896-6273(00)80877-6

    • PubMed
    • Google Scholar
20. 1. Hsu AI
    2. Yttri EA

    (2021) B-SOiD, an open-source unsupervised algorithm for identification and fast prediction of behaviors

    Nature Communications 12:5188.

    https://doi.org/10.1038/s41467-021-25420-x

    • PubMed
    • Google Scholar
21. Preprint

    1. Joyce M
    2. Falconio FA
    3. Blackhurst L
    4. Prieto-Godino L
    5. French AS
    6. Gilestro GF

    (2023) Divergent Evolution of Sleep Functions

    bioRxiv.

    https://doi.org/10.1101/2023.05.27.541573

    • Google Scholar
22. 1. Kabra M
    2. Robie AA
    3. Rivera-Alba M
    4. Branson S
    5. Branson K

    (2013) JAABA: interactive machine learning for automatic annotation of animal behavior

    Nature Methods 10:64–67.

    https://doi.org/10.1038/nmeth.2281

    • PubMed
    • Google Scholar
23. 1. Kaduskar B
    2. Kushwah RBS
    3. Auradkar A
    4. Guichard A
    5. Li M
    6. Bennett JB
    7. Julio AHF
    8. Marshall JM
    9. Montell C
    10. Bier E

    (2022) Reversing insecticide resistance with allelic-drive in Drosophila melanogaster

    Nature Communications 13:291.

    https://doi.org/10.1038/s41467-021-27654-1

    • PubMed
    • Google Scholar
24. 1. Lubba CH
    2. Sethi SS
    3. Knaute P
    4. Schultz SR
    5. Fulcher BD
    6. Jones NS

    (2019) catch22: CAnonical Time-series CHaracteristics

    Data Mining and Knowledge Discovery 33:1821–1852.

    https://doi.org/10.1007/s10618-019-00647-x

    • Google Scholar
25. 1. Mathis A
    2. Mamidanna P
    3. Cury KM
    4. Abe T
    5. Murthy VN
    6. Mathis MW
    7. Bethge M

    (2018) DeepLabCut: markerless pose estimation of user-defined body parts with deep learning

    Nature Neuroscience 21:1281–1289.

    https://doi.org/10.1038/s41593-018-0209-y

    • PubMed
    • Google Scholar
26. 1. McDermott-Rouse A
    2. Minga E
    3. Barlow I
    4. Feriani L
    5. Harlow PH
    6. Flemming AJ
    7. Brown AEX

    (2021) Behavioral fingerprints predict insecticide and anthelmintic mode of action

    Molecular Systems Biology 17:e10267.

    https://doi.org/10.15252/msb.202110267

    • PubMed
    • Google Scholar
27. 1. Nath T
    2. Mathis A
    3. Chen AC
    4. Patel A
    5. Bethge M
    6. Mathis MW

    (2019) Using DeepLabCut for 3D markerless pose estimation across species and behaviors

    Nature Protocols 14:2152–2176.

    https://doi.org/10.1038/s41596-019-0176-0

    • PubMed
    • Google Scholar
28. 1. Pereira TD
    2. Aldarondo DE
    3. Willmore L
    4. Kislin M
    5. Wang SS-H
    6. Murthy M
    7. Shaevitz JW

    (2019) Fast animal pose estimation using deep neural networks

    Nature Methods 16:117–125.

    https://doi.org/10.1038/s41592-018-0234-5

    • PubMed
    • Google Scholar
29. 1. Remnant EJ
    2. Morton CJ
    3. Daborn PJ
    4. Lumb C
    5. Yang YT
    6. Ng HL
    7. Parker MW
    8. Batterham P

    (2014) The role of Rdl in resistance to phenylpyrazoles in Drosophila melanogaster

    Insect Biochemistry and Molecular Biology 54:11–21.

    https://doi.org/10.1016/j.ibmb.2014.08.008

    • PubMed
    • Google Scholar
30. 1. Samantsidis G-R
    2. Panteleri R
    3. Denecke S
    4. Kounadi S
    5. Christou I
    6. Nauen R
    7. Douris V
    8. Vontas J

    (2020) What I cannot create, I do not understand’: functionally validated synergism of metabolic and target site insecticide resistance

    Proceedings of the Royal Society B 287:20200838.

    https://doi.org/10.1098/rspb.2020.0838

    • Google Scholar
31. 1. Shaw PJ
    2. Cirelli C
    3. Greenspan RJ
    4. Tononi G

    (2000) Correlates of sleep and waking in Drosophila melanogaster

    Science 287:1834–1837.

    https://doi.org/10.1126/science.287.5459.1834

    • PubMed
    • Google Scholar
32. 1. Simon JC
    2. Dickinson MH

    (2010) A new chamber for studying the behavior of Drosophila

    PLOS ONE 5:e8793.

    https://doi.org/10.1371/journal.pone.0008793

    • PubMed
    • Google Scholar
33. 1. Wiggin TD
    2. Goodwin PR
    3. Donelson NC
    4. Liu C
    5. Trinh K
    6. Sanyal S
    7. Griffith LC

    (2020) Covert sleep-related biological processes are revealed by probabilistic analysis in Drosophila

    PNAS 117:10024–10034.

    https://doi.org/10.1073/pnas.1917573117

    • PubMed
    • Google Scholar
34. 1. Wiltschko AB
    2. Johnson MJ
    3. Iurilli G
    4. Peterson RE
    5. Katon JM
    6. Pashkovski SL
    7. Abraira VE
    8. Adams RP
    9. Datta SR

    (2015) Mapping sub-second structure in mouse behavior

    Neuron 88:1121–1135.

    https://doi.org/10.1016/j.neuron.2015.11.031

    • PubMed
    • Google Scholar
35. 1. Wiltschko AB
    2. Tsukahara T
    3. Zeine A
    4. Anyoha R
    5. Gillis WF
    6. Markowitz JE
    7. Peterson RE
    8. Katon J
    9. Johnson MJ
    10. Datta SR

    (2020) Revealing the structure of pharmacobehavioral space through motion sequencing

    Nature Neuroscience 23:1433–1443.

    https://doi.org/10.1038/s41593-020-00706-3

    • PubMed
    • Google Scholar

Author details

1. Hannah Jones

   Department of Life Sciences, Imperial College London, London, United Kingdom

   Contribution

   Data curation, Investigation, Methodology, Project administration

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-9481-8094

2. Jenny A Willis

   Syngenta, Jealott’s Hill International Research Centre, Bracknell, United Kingdom

   Contribution

   Resources, Validation

   Competing interests

   No competing interests declared

3. Lucy C Firth

   Syngenta, Jealott’s Hill International Research Centre, Bracknell, United Kingdom

   Contribution

   Resources, Validation

   Competing interests

   No competing interests declared

4. Carlo NG Giachello

   Syngenta, Jealott’s Hill International Research Centre, Bracknell, United Kingdom

   Contribution

   Methodology

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-4186-1571

5. Giorgio F Gilestro

   Department of Life Sciences, Imperial College London, London, United Kingdom

   Contribution

   Conceptualization, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Project administration, Writing – review and editing

   For correspondence

   giorgio@gilest.ro

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7512-8541

• 1,049

  views

• 64

  downloads

• 2

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.