Mapping model units to visual neurons reveals population code for social behaviour
Open accessThe DOI matches its Crossref record; 99 of the 99 references it cites were found in a registry or an index; 34 of the 58 sources its claims cite were checked for support.
Abstract
The rich variety of behaviours observed in animals arises through the interplay between sensory processing and motor control. To understand these sensorimotor transformations, it is useful to build models that predict not only neural responses to sensory input 1–5 but also how each neuron causally contributes to behaviour 6,7 . Here we demonstrate a novel modelling approach to identify a one-to-one mapping between internal units in a deep neural network and real neurons by predicting the behavioural changes that arise from systematic perturbations of more than a dozen neuronal cell types. A key ingredient that we introduce is ‘knockout training’, which involves perturbing the network during training to match the perturbations of the real neurons during behavioural experiments. We apply this approach to model the sensorimotor transformations of Drosophila melanogaster males during a complex, visually guided social behaviour 8–11 . The visual projection neurons at the interface between the optic lobe and central brain form a set of discrete channels 12 , and prior work indicates that each channel encodes a specific visual feature to drive a particular behaviour 13,14 . Our model reaches a different conclusion: combinations of visual projection neurons, including those involved in non-social behaviours, drive male interactions with the female, forming a rich population code for behaviour. Overall, our framework consolidates behavioural effects elicited from various neural perturbations into a single, unified model, providing a map from stimulus to neuronal cell type to behaviour, and enabling future incorporation of wiring diagrams of the brain 15 into the model.
Overview
CoChat Summary
## TL;DR
The paper introduces “knockout training,” a DNN training scheme that matches network perturbations to neuronal silencing, to map model units one-to-one onto lobula columnar (LC/LPLC) neuron types in male *Drosophila*. The resulting model suggests courtship behavior is driven by a distributed population code: multiple LC types jointly encode female visual features and jointly contribute to singing and pursuit.
## Key findings
- The 1-to-1 network predicted real LC responses to artificial stimuli and naturalistic stimuli, with an average noise-corrected \(R^2 \approx 0.35\) for held-out neural responses.
- When a linear mapping from all model LC units to one real LC type was fit using neural data, held-out prediction improved to a noise-corrected \(R^2 \approx 0.65\).
- For behavior, cumulative inactivation showed that most model LC units contributed to multiple outputs, and each behavioral output was driven by multiple LC units.
- Connectome analysis found broad sharing of inputs and outputs: 60.2% of presynaptic neuron types projected to 2 or more LC types, and 55.6% of downstream neuron types received input from 2 or more LC types.
## Methods
- Built a 3-part DNN: a vision network, a 23-unit LC bottleneck, and a decision network predicting six outputs (forward, lateral, angular velocity; sine, Pfast, and Pslow song) from 10-frame visual input sequences.
- Trained on courtship data from 459 male-female pairs, including 23 LC silencing conditions plus control; behavior was extracted from video and audio, and training used about 11.6 million samples.
- Used “knockout training” by zeroing the model unit corresponding to the silenced LC type during training; compared against dropout, no-knockout, and untrained baselines.
- Validated neural predictions with two-photon calcium imaging from 5 LC types (LC6, LC11, LC12, LC15, LC17) and analyzed LC tuning plus causal contribution to behavior with cumulative inactivation.
## Limitations
- The paper states that if silencing an LC type does not noticeably change behavior, the model cannot infer that LC type’s tuning.
- It notes mismatches between real LC responses and model predictions, possibly because training data came from freely moving males while imaging was in head-fixed flies passively viewing stimuli.
- The text makes evident that the connectome analysis is based on a female fly connectome, not a male one, so inference about male circuits is indirect (text-evident limitation).
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures


![Fig. 3 Visual features of female motion are distributed across the population of model LC units. a, Almost all model LC units responded to a fictive female changing in size, position and rotation. b, Cross-validated R2 between each primary visual parameter and model LC responses for natural stimulus sequences. Columns are sorted based on female size (top). The end column of each row (all) is the cross-validated R2 between a linear combination (identified via ridge regression) between all model LC units and a single visual parameter. c, We characterized the tuning preferences of each model LC unit by systematically varying the three visual parameters and computing a heat map of the model LC responses. Each input sequence was static (that is, all ten frames were repeats of the same image). d, Tuning heat maps for example model LC units (see Extended Data Fig. 9 for all LC types). e, We used variance decomposition (Methods) to decompose the response variance V[y] of each model LC unit into components solely due to either female size V[y¯size] (blue), position V[y¯position] (orange) or rotation V[y¯rotation] (green) as well as interactions between these visual parameters V[yinteractions] (black). A large fraction of response variance for a given parameter indicates that a model LC unit more strongly changes its response y to variations in this parameter relative to those in other parameters. Because the 1-to-1 network is deterministic, all response variance can be attributed to variations of the parameters (that is, there is no repeat-to-repeat variability). f, Example model LC responses to dynamic stimulus sequences in which the fictive female solely varied either her size, position or rotation angle over time (dashed traces). Different model LC units appear either to directly encode a visual parameter (for example, LC10a encodes size) or encode features derived from the parameter, such as a delay (LC17, arrows) or speed at which female size changes (LC13). Responses for all model LC units are in Extended Data Fig. 10. g, R2 between model responses and visual parameter features for the stimulus sequences in f. Columns are in the same order as those in b. Source data](https://figures.cochat.ai/paper-figures/5812d375574849bfa030a977e2b30f25/mapping-model-units-to-visual-neurons-reveals-population-code-for-social-figure-3.jpg)














Verification
Claims the paper makes
17 Valid · 1 Not valid · 7 Not checked · 114 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“we found that silencing LC10a neurons resulted in failures to initiate chasing, as male-to-female distances remained large over time (Fig. 1d, middle, 1e, top)”
§ Mapping model units to visual neurons reveals population code for social behaviour > A deep network model of vision to behaviour The Drosophila visual system contains a bottleneck between the optic lobes and the central brain in the form of visual projection neurons, which comprise approximately 200 different cell types18,19. The primary cell types of this bottleneck (Fig. 1a) are the 57 lobula columnar (LC) and lobula plate (LPLC) neuron types identified so far (we use ‘LC types’ to refer both to LC and LPLC neuron types), making up about 3.5% of all neurons in the brain. The LC neuron types receive input from the lobula and lobula plate in the optic lobe and send axons to optic glomeruli in the central brain12,20. Neurons of a single LC type innervate only one optic glomerulus in the posterior lateral protocerebrum, posterior ventrolateral protocerebrum or anterior optic tubercle neuropils, and prior studies have uncovered mappings between specific LC types, visual features and specific behaviours11,21–28. For example, LPLC2 neurons respond to a looming object and synapse onto the giant fibre neuron to drive an escape take-off25. LC11 neurons respond to small, moving spots and contribute to freezing behaviour27,28. For courtship, the LC10a neurons (and LC9 neurons, to a lesser extent) of a male participate in tracking the position of the female and driving turns towards the female11,22,23, but it is not yet known whether other LC types contribute to male social behaviours. As recordings from LC neurons reveal that even simple stimuli can drive responses in multiple LC types29–31, we explored whether the representation of the female during courtship might be distributed across the LC population, and similarly whether multiple LC types might be required to drive behaviour. We designed a novel DNN modelling approach for identifying the functional roles of LC neuron types using behavioural data from genetically altered flies. The DNN model has three components: (1) a front-end convolutional vision network that reflects processing in the optic lobe; (2) a bottleneck layer of LC units in which each model LC unit represents the summed activity of neurons of the same LC type (that is, the overall activity level of an optic glomerulus); and (3) a decision network with dense connections that maps LC responses to behaviour, reflecting downstream processing in the central brain and ventral nerve cord (Fig. 1a). We imposed the bottleneck layer to have the same number of units as LC neuron types we manipulated, and our goal was to identify a one-to-one mapping between model LC units and LC neuron types. We did not incorporate biological realism into the vision and decision networks, opting instead for highly expressive mappings to ensure accurate prediction; we focused on explaining LC function. We collected training data to fit the model by blocking synaptic transmission32 in each of 23 different LC types in male flies12,33 and recorded the movements of the LC-silenced male and song production during natural courtship (Methods). We then devised a fitting procedure called knockout training, which involves training the model using the entire behavioural dataset of both perturbed and unperturbed sets of males. Critically, when training the model on data from a male with a particular LC type silenced (Fig. 1b), we set to 0 (that is, we knocked out) the activity of the corresponding model LC unit (correspondence was arbitrarily chosen at initialization; see Methods). The resulting model captures the behavioural repertoire of each genetically altered fly when the corresponding model LC unit is silenced, thereby aligning the model LC units to the real LC neurons. In simulations (Extended Data Fig. 2), knockout training correctly identified the activity and contribution to behaviour of each silenced neuron type (a one-to-one mapping) for neuron types that, when silenced, led to changes in behaviour. We refer to the resulting DNN model as the ‘1-to-1 network’. Before fitting the model with courtship data (Fig. 1c), we quantified the extent to which (bilaterally) silencing each LC neuron type changes behaviour of the male fly (Extended Data Fig. 1). Consistent with previous studies11,23, we found that silencing LC10a neurons resulted in failures to initiate chasing, as male-to-female distances remained large over time (Fig. 1d, middle, 1e, top); we found similar results with silencing LC922. We also found strong effects on both chasing and singing when silencing other LC types. For example, silencing LC6 and LC26 neurons resulted in stronger and more persistent chasing, as male-to-female distances remained small over time (Fig. 1d, bottom, 1e, top). We observed a large number of LC types (LC4, LC6, LC11, LPLC1, LPLC2 and LC10bc) that, after silencing, significantly increased the amount of sine song relative to pulse song (Fig. 1e, bottom)—sine song typically occurs near the female34. Across behavioural measures, we found that the silencing of any single LC type did not match the behavioural deficits of blind flies (Extended Data Fig. 1). This suggests that many LC types would need to be silenced together to uncover large effects on courtship. We therefore modelled the perturbed behavioural data with the 1-to-1 network, enabling us to silence any possible combination of LC types in silico. We performed knockout training to fit the parameters of the 1-to-1 network. The model inputs were videos of the visual input of the male fly during natural courtship (Methods and Fig. 1f, left); the model outputs comprised the male movements (forward, lateral and angular velocity) and song production, which included sine song and two forms of pulse song (Pfast and Pslow35). The 1-to-1 network reliably predicted these behavioural variables in held-out data (Fig. 1f, right and Extended Data Fig. 3). Notably, the 1-to-1 network also predicted differences in behaviour observed across silenced LC types (Fig. 1g,h and Extended Data Fig. 4). We confirmed that knockout training outperformed other possible training procedures, such as dropout training36 and training without knockout (that is, an unconstrained network) (Extended Data Figs. 3 and 4), and that results were largely consistent for different random initializations of the 1-to-1 network (Extended Data Fig. 5). Thus, the 1-to-1 network reliably estimated the behaviour of the male from visual input alone, even for male flies with a silenced LC type.
What the sources say
- Partly supportedcites [11] Hindmarsh Sten, T., Li, R., Otopalik, A. & Ruta, V. Sexual arousal gates visual processing during drosophila courtship. Nature 595, 549–553 (2021). sourceExperimental — accuracy not yet measured
The 'LC10a gain is modulated during courtship' section reports that silencing LC10a neurons attenuated turning toward a visual target during tethered courtship. The Extended Data Figure 7 caption likewise reports failures to turn toward the silenced side, but neither passage reports failures to initiate chasing or persistently large male-to-female distances after silencing. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites [23] Ribeiro, I. M. et al. Visual projection neurons mediating directed courtship in Drosophila. Cell 174, 607–621 (2018). source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“A shortcoming of this approach is that the DNN does not predict how an individual neuron causally contributes to behaviour, making it difficult to interpret the role of the neuron in the sensorimotor transformation.”
§ Mapping model units to visual neurons reveals population code for social behaviour > Main
What the sources say
- Not checked — no full text of the sourcecites [1] Yamins, D. L. & DiCarlo, J. J. Using goal-driven deep learning models to understand sensory cortex. Nat. Neurosci. 19, 356–365 (2016). source
- Partly supportedcites [2] Sussillo, D., Churchland, M. M., Kaufman, M. T. & Shenoy, K. V. A neural network that finds a naturalistic solution for the production of muscle activity. Nat. Neurosci. 18, 1025–1033 (2015). sourceExperimental — accuracy not yet measured
The Discussion (page unavailable) says individual-neuron responses rarely resemble muscle activity and warns against interpreting each response as a representation of a meaningful variable. That supports the difficulty of interpreting a neuron's role, but the paper does not say the network cannot predict an individual neuron's causal contribution to behaviour. (LLM, conf=0.84)
- Not checked — no full text of the sourcecites [3] Richards, B. A. et al. A deep learning framework for neuroscience. Nat. Neurosci. 22, 1761–1770 (2019). source
- Could not verifycites [5] Mano, O., Creamer, M. S., Badwan, B. A. & Clark, D. A. Predicting individual neuron responses with anatomically constrained task optimization. Curr. Biol. 31, 4062–4075 (2021). sourceExperimental — accuracy not yet measured
The Introduction and Discussion describe models that predict motion and reproduce properties of individual neurons, but do not address whether the models predict a neuron's causal contribution to behaviour. Searches for causal contribution, behavioural effects, and the neuron's role in sensorimotor transformation found no passage that settles the claim. (LLM, conf=0.73)
- Could not verifycites [16] Michaels, J. A., Schaffelhofer, S., Agudelo-Toro, A. & Scherberger, H. A goal-driven modular neural network predicts parietofrontal neural dynamics during grasping. Proc. Natl Acad. Sci. USA 117, 32124–32135 (2020). sourceExperimental — accuracy not yet measured
Across five targeted searches (individual neuron / causal contribution to behaviour, shortcomings or limitations of the DNN or goal-driven approach, interpretability and black-box issues, single-unit vs distributed coding), no passage of our copy states or discusses this limitation of deep-network modelling. The closest material runs the other way at the population level — the Results ('Modular Mechanisms of Memory and Movement Execution within the mRNN Model') stress that 'since our models are fully observable, we were able to probe our mRNN networks for explanations of the computations necessary for the task', and the lesioning analysis (Results, Fig. 6) silences random subsets of neurons per module, not individual neurons — but none of this asserts, or contradicts, the specific claim that a DNN fails to predict how an individual neuron causally contributes to behaviour. The claim therefore cannot be checked against this source. (LLM, conf=0.62)
- Partly supportedcites [17] Zhou, B., Li, Z., Kim, S., Lafferty, J. & Clark, D. A. Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons. eLife 11, e72067 (2022). sourceExperimental — accuracy not yet measured
The Results section (“An anatomically constrained mathematical model”; page unavailable) says the network combines model-unit responses to predict collision probability, not behaviour. The Discussion (page unavailable) says individual LPLC2 responses should be interpreted in light of the population response. These passages support a limitation on interpreting individual neurons, but do not establish the specific claim that the DNN cannot predict an individual neuron's causal contribution to behaviour. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“For courtship, the LC10a neurons (and LC9 neurons, to a lesser extent) of a male participate in tracking the position of the female and driving turns towards the female11,22,23”
§ Mapping model units to visual neurons reveals population code for social behaviour > A deep network model of vision to behaviour The Drosophila visual system contains a bottleneck between the optic lobes and the central brain in the form of visual projection neurons, which comprise approximately 200 different cell types18,19. The primary cell types of this bottleneck (Fig. 1a) are the 57 lobula columnar (LC) and lobula plate (LPLC) neuron types identified so far (we use ‘LC types’ to refer both to LC and LPLC neuron types), making up about 3.5% of all neurons in the brain. The LC neuron types receive input from the lobula and lobula plate in the optic lobe and send axons to optic glomeruli in the central brain12,20. Neurons of a single LC type innervate only one optic glomerulus in the posterior lateral protocerebrum, posterior ventrolateral protocerebrum or anterior optic tubercle neuropils, and prior studies have uncovered mappings between specific LC types, visual features and specific behaviours11,21–28. For example, LPLC2 neurons respond to a looming object and synapse onto the giant fibre neuron to drive an escape take-off25. LC11 neurons respond to small, moving spots and contribute to freezing behaviour27,28. For courtship, the LC10a neurons (and LC9 neurons, to a lesser extent) of a male participate in tracking the position of the female and driving turns towards the female11,22,23, but it is not yet known whether other LC types contribute to male social behaviours. As recordings from LC neurons reveal that even simple stimuli can drive responses in multiple LC types29–31, we explored whether the representation of the female during courtship might be distributed across the LC population, and similarly whether multiple LC types might be required to drive behaviour. We designed a novel DNN modelling approach for identifying the functional roles of LC neuron types using behavioural data from genetically altered flies. The DNN model has three components: (1) a front-end convolutional vision network that reflects processing in the optic lobe; (2) a bottleneck layer of LC units in which each model LC unit represents the summed activity of neurons of the same LC type (that is, the overall activity level of an optic glomerulus); and (3) a decision network with dense connections that maps LC responses to behaviour, reflecting downstream processing in the central brain and ventral nerve cord (Fig. 1a). We imposed the bottleneck layer to have the same number of units as LC neuron types we manipulated, and our goal was to identify a one-to-one mapping between model LC units and LC neuron types. We did not incorporate biological realism into the vision and decision networks, opting instead for highly expressive mappings to ensure accurate prediction; we focused on explaining LC function. We collected training data to fit the model by blocking synaptic transmission32 in each of 23 different LC types in male flies12,33 and recorded the movements of the LC-silenced male and song production during natural courtship (Methods). We then devised a fitting procedure called knockout training, which involves training the model using the entire behavioural dataset of both perturbed and unperturbed sets of males. Critically, when training the model on data from a male with a particular LC type silenced (Fig. 1b), we set to 0 (that is, we knocked out) the activity of the corresponding model LC unit (correspondence was arbitrarily chosen at initialization; see Methods). The resulting model captures the behavioural repertoire of each genetically altered fly when the corresponding model LC unit is silenced, thereby aligning the model LC units to the real LC neurons. In simulations (Extended Data Fig. 2), knockout training correctly identified the activity and contribution to behaviour of each silenced neuron type (a one-to-one mapping) for neuron types that, when silenced, led to changes in behaviour. We refer to the resulting DNN model as the ‘1-to-1 network’. Before fitting the model with courtship data (Fig. 1c), we quantified the extent to which (bilaterally) silencing each LC neuron type changes behaviour of the male fly (Extended Data Fig. 1). Consistent with previous studies11,23, we found that silencing LC10a neurons resulted in failures to initiate chasing, as male-to-female distances remained large over time (Fig. 1d, middle, 1e, top); we found similar results with silencing LC922. We also found strong effects on both chasing and singing when silencing other LC types. For example, silencing LC6 and LC26 neurons resulted in stronger and more persistent chasing, as male-to-female distances remained small over time (Fig. 1d, bottom, 1e, top). We observed a large number of LC types (LC4, LC6, LC11, LPLC1, LPLC2 and LC10bc) that, after silencing, significantly increased the amount of sine song relative to pulse song (Fig. 1e, bottom)—sine song typically occurs near the female34. Across behavioural measures, we found that the silencing of any single LC type did not match the behavioural deficits of blind flies (Extended Data Fig. 1). This suggests that many LC types would need to be silenced together to uncover large effects on courtship. We therefore modelled the perturbed behavioural data with the 1-to-1 network, enabling us to silence any possible combination of LC types in silico. We performed knockout training to fit the parameters of the 1-to-1 network. The model inputs were videos of the visual input of the male fly during natural courtship (Methods and Fig. 1f, left); the model outputs comprised the male movements (forward, lateral and angular velocity) and song production, which included sine song and two forms of pulse song (Pfast and Pslow35). The 1-to-1 network reliably predicted these behavioural variables in held-out data (Fig. 1f, right and Extended Data Fig. 3). Notably, the 1-to-1 network also predicted differences in behaviour observed across silenced LC types (Fig. 1g,h and Extended Data Fig. 4). We confirmed that knockout training outperformed other possible training procedures, such as dropout training36 and training without knockout (that is, an unconstrained network) (Extended Data Figs. 3 and 4), and that results were largely consistent for different random initializations of the 1-to-1 network (Extended Data Fig. 5). Thus, the 1-to-1 network reliably estimated the behaviour of the male from visual input alone, even for male flies with a silenced LC type.
What the sources say
- Partly supportedcites [11] Hindmarsh Sten, T., Li, R., Otopalik, A. & Ruta, V. Sexual arousal gates visual processing during drosophila courtship. Nature 595, 549–553 (2021). sourceExperimental — accuracy not yet measured
The section ‘LC10a gain is modulated during courtship’ reports that silencing LC10a attenuated males’ turns toward a visual target, while activating it drove ipsilateral turning; it also describes LC10a as essential for accurate courtship tracking. The passages do not establish the claim’s additional point that LC9 neurons participate to a lesser extent (searched: LC9). (LLM, conf=0.88)
- Partly supportedcites [22] Bidaye, S. S. et al. Two brain pathways initiate distinct forward walking programs in Drosophila. Neuron 108, 469–485 (2020). sourceExperimental — accuracy not yet measured
The Results section on courtship reports that silencing LC9 decreased males’ following of females. The Discussion says LC10 neurons participate in object tracking during courtship, but does not specify LC10a or establish that either LC10a or LC9 drives turns toward the female; the Results attribute initiated forward turns to P9. No page numbers are available in our copy. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites [23] Ribeiro, I. M. et al. Visual projection neurons mediating directed courtship in Drosophila. Cell 174, 607–621 (2018). source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“exact matches were unlikely owing to differences in behavioural state during courtship (on which the 1-to-1 network was trained) and during imaging11,31.”
§ Mapping model units to visual neurons reveals population code for social behaviour > Comparing real and model neural activity
What the sources say
- Partly supportedcites [11] Hindmarsh Sten, T., Li, R., Otopalik, A. & Ruta, V. Sexual arousal gates visual processing during drosophila courtship. Nature 595, 549–553 (2021). sourceExperimental — accuracy not yet measured
The section “LC10a gain is modulated during courtship” (page unavailable) reports weak LC10a responses in unaroused males and stronger responses during courtship, supporting the point that behavioural state affects LC responses. It does not say that the model and imaged responses were unlikely to match exactly because the animals were in different states during training and imaging; the paper also describes imaging during courtship. (LLM, conf=0.80)
- Partly supportedcites [31] Turner, M. H., Krieger, A., Pang, M. M. & Clandinin, T. R. Visual and motor signatures of locomotion dynamically shape a population code for feature detection in drosophila. eLife 11, e82587 (2022). sourceExperimental — accuracy not yet measured
The Results section, “Walking behavior selectively suppresses responses of small-object detecting glomeruli” (page unavailable), shows that walking weakens visual responses in some LC-associated glomeruli. This supports the premise that behavioural state can affect LC responses, but the paper does not compare courtship with imaging or establish that such a difference makes exact model–response matches unlikely. (LLM, conf=0.83)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Model LC10a was driven by female position (in front of the male), consistent with prior work11,23, but we found this was only true for conditions in which he is close and directly behind her (Fig. 3d, bottom).”
§ Mapping model units to visual neurons reveals population code for social behaviour > Visual feature encoding of the model LC units
What the sources say
- Supportedcites [11] Hindmarsh Sten, T., Li, R., Otopalik, A. & Ruta, V. Sexual arousal gates visual processing during drosophila courtship. Nature 595, 549–553 (2021). source
The citation is attached to the prior-work point that LC10a responds to a target in front of the male, not to the citing authors’ subsequent close-and-directly-behind condition. In “LC10a gain is modulated during courtship” (page unavailable), the source reports robust LC10a activation as a visual target sweeps across the male’s ipsilateral hemifield; its Methods specify that stimuli were presented in front of the male. (LLM, conf=0.88)
- Not checked — no full text of the sourcecites [23] Ribeiro, I. M. et al. Visual projection neurons mediating directed courtship in Drosophila. Cell 174, 607–621 (2018). source
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
90 more claims · 6 with their cited sources checked
Create a free account to see all 139 claims and to verify the 3 references still open.
- Probably not a substantial claim. 33
- The quote only partly states this claim. 1
- It may state two separate claims as one. 15
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Connectome-constrained DMN models generate meaningful predictions even in the complete absence of neural activity measurements, but can be further constrained by sparse measurements of neural activity as we showed (Fig. 3), or indeed directly fitted to measured neural activity51 and behaviour54.”
From Connectome-constrained networks predict neural activity across the fly visual system (2024) · cites this paper as [54]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“This adhesive force increases the normal force toward the object and the frictional force (in insects, normal and frictional forces are also correlated 52 ).”
From NeuroMechFly v2: simulating embodied sensorimotor control in adult Drosophila (2024) · cites this paper as [52]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“We opted for a relatively high adhesion force (20-40 mN) to be consistent with the high, but somewhat variable, forces reported across insect species 52,51 .”
From NeuroMechFly v2: simulating embodied sensorimotor control in adult Drosophila (2024) · cites this paper as [52]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Substantial normal forces (10-100 times body weight 51,52,53 ) and frictional forces emerge from interactions between the substrate and the adhesive pad 49,54 .”
From NeuroMechFly v2: simulating embodied sensorimotor control in adult Drosophila (2024) · cites this paper as [52]
Experimental — accuracy not yet measured
Citations
Cites
(2014) · Cited by 33,419
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818 · cited 1×
“3), overall mean changes to behaviour across silenced LC types (Extended Data Fig.”
MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., & Adam, H. (2017) · arXiv · doi:10.48550/arxiv.1704.04861 · Cited by 19,015
TensorFlow: A system for large-scale machine learning
Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghemawat, S., Irving, G., Isard, M., et al. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.08695 · Cited by 11,822
Emergence of simple-cell receptive field properties by learning a sparse code for natural images
Olshausen, B. A., & Field, D. J. (1996) · Nature · doi:10.1038/381607a0 · Cited by 5,694
Random Features for Large-Scale Kernel Machines
Rahimi, A., & Recht, B. (2007) · Cited by 2,783
Shadlen, M. N., & Newsome, W. T. (1998) · The Journal of Neuroscience · doi:10.1523/jneurosci.18-10-03870.1998 · Cited by 2,178
Performance-optimized hierarchical models predict neural responses in higher visual cortex
Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., & DiCarlo, J. J. (2014) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1403112111 · Cited by 1,924
Context-dependent computation by recurrent dynamics in prefrontal cortex
Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013) · Nature · doi:10.1038/nature12742 · Cited by 1,847
The importance of mixed selectivity in complex cognitive tasks
Rigotti, M., Barak, O., Warden, M. R., Wang, X.-J., Daw, N. D., Miller, E. K., & Fusi, S. (2013) · Nature · doi:10.1038/nature12160 · Cited by 1,639
Research works printed but not linked: 1
- Cadena, S. A. et al. How well do deep neural networks trained on object recognition characterize the mouse visual system? In Real Neurons & Hidden Units: Future Directions at the Intersection of Neuroscience and Artificial Intelligence@ NeurIPS 2019 (2019).
Cited by
Connectome-constrained networks predict neural activity across the fly visual system
Lappalainen, J. K., Tschopp, F. D., Prakhya, S., McGill, M., Nern, A., Shinomiya, K., Takemura, S.-Y., Gruntman, E., Macke, J. H., & Turaga, S. C. (2024) · Nature · doi:10.1038/s41586-024-07939-3 · Cited by 128
“3), or indeed directly fitted to measured neural activity 51 and behaviour 54 .”
BackgroundA Drosophila computational brain model reveals sensorimotor processing
Shiu, P. K., Sterne, G. R., Spiller, N., Franconville, R., Sandoval, A., Zhou, J., Simha, N., Kang, C. H., Yu, S., Kim, J. S., et al. (2024) · Nature · doi:10.1038/s41586-024-07763-9 · Cited by 97
“A second approach is to train a deep neural network to perform a task, and measure how neurons in that model compare with actual neurons in the brain 15,48 .”
BackgroundArtificial intelligence for life sciences: A comprehensive guide and future trends
Luo, M., Yang, W., Bai, L., Zhang, L., Huang, J., Cao, Y.-H., Xie, Y., Tong, L., Zhang, H., Lei, Y., et al. (2024) · The Innovation Life · doi:10.59717/j.xinn-life.2024.100105 · Cited by 61
The fly connectome reveals a path to the effectome
Pospisil, D. A., Aragon, M. J., Dorkenwald, S., Matsliah, A., Sterling, A. R., Schlegel, P., Yu, S.-C., McKellar, C. E., Costa, M., Eichler, K., et al. (2024) · Nature · doi:10.1038/s41586-024-07982-0 · Cited by 36
“32,33).”
Self-citationNeuroMechFly v2: simulating embodied sensorimotor control in adult Drosophila
Wang-Chen, S., Stimpfling, V. A., Lam, T. K. C., Özdil, P. G., Genoud, L., Hurtak, F., & Ramdya, P. (2024) · Nature Methods · doi:10.1038/s41592-024-02497-y · Cited by 43
“We obtained similar results even when using a smaller subset of neurons that provide inputs to LC9 and LC10–LC neurons implicated in courtship in particular 52 (Extended Data Fig.”
Metadata
- Authors
- Cowley, B. R., Calhoun, A. J., Rangarajan, N., Ireland, E., Turner, M. H., Pillow, J. W., & Murthy, M.
- Affiliations
- Princeton University (US)
- Stanford Medicine (US)
- Corresponding author
- Cowley, Benjamin R.
- Murthy, Mala
- ORCID
- Venue
- Nature
- Published
- 2024-05-22
- Volume
- 629
- Issue
- 8014
- Pages
- 1100-1108
- ISSN
- 0028-0836 (ISSN-L)
- 1476-4687
- PMC
- PMC11136655
- OpenAlex
- W4398201937
- Cited by
- 55
- References
- 64
- Citations by year
- 2026: 25 · 2025: 10 · 2024: 18
- FWCI
- 7.57
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Nature (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (submittedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Cold Spring Harbor Laboratory Institutional Repository (Cold Spring Harbor Laboratory) (acceptedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Funders
- Howard Hughes Medical Institute
- National Institutes of Health (NIH-P40OD018537, P40OD018537)
- National Institute of Mental Health (RF1MH129268)
- National Institute on Drug Abuse (RF1DA056404)
- National Institute of Neurological Disorders and Stroke (R01NS110060, R01NS104899, R35NS111580)
- Topics
- Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurobiology and Insect Physiology Research
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
- Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
- Keywords
- population coding, social behavior, Drosophila melanogaster, sensorimotor transformation, deep neural network, behavior prediction
- MeSH terms
- Animals, Brain, Drosophila melanogaster, Female, Male, Models, Neurological, Nerve Net, Neurons, Optic Lobe, Nonmammalian, Social Behavior, Visual Perception
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Mapping model units to visual neurons reveals population code for social behaviour. https://app.cochat.ai/papers/10.1038/s41586-024-07451-8
- The paper
- Cowley, B. R., Calhoun, A. J., Rangarajan, N., Ireland, E., Turner, M. H., Pillow, J. W., & Murthy, M. (2024). Mapping model units to visual neurons reveals population code for social behaviour. Nature. https://doi.org/10.1038/s41586-024-07451-8
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.