README: Dataset for Alcacer et al., Cell Reports 2025

--------------------
GENERAL INFORMATION 
--------------------
Generated on: 2025-05-27
Author: Cristina Alcacer
Email: cristina.alcacer81@gmail.com
ORCID: 0000-0003-2502-0983
Dataset Title: Dataset for Alcacer et al., Cell Reports 2025 Figures
Description: Tabular dataset associated with each figure panel in the publication, supporting analysis of behavioral and neural activity in dyskinetic and control mice.
Data Collection Period: 2017-08-30 to 2021-06-04
Software Dependencies: Microsoft Excel

----------------------------------------
DATA SPECIFIC INFORMATION FOR FIGURE 1 
----------------------------------------
Figure title: Behavioral changes in dyskinetic mice revealed by unsupervised behavioral clustering.
Date of Creation: 2025-05-27
Panels Covered: Figure 1B, 1F, 1I and 1J.

Figure 1B: Time Moving and Acceleration When Moving
- Variables (2):
	1. Time moving (%)
	Description: Percentage of session time mice spent moving, based on a 0.05 g BA 	threshold derived from IMU recordings (Figure S2C). 	
	2. Acceleration when moving (g)
	Description: Average acceleration (BA) during active movement periods.
- Units of measurement: % and g
- Number of Rows: 51
- Column Names: 
	A. Mouse ID
	B. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	C. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D. Time Moving (%)
	E. Acceleration When Moving (g)

Figure 1F: ROC Analysis for Clustering Accuracy
- Variables (3):
	1. True positive rate 
	Description: Proportion of correctly identified same-cluster behavioral segments.
	2. False positive rate
	Description: Proportion of misclassified same-cluster segments.
	3. Balanced Accuracy (BACC) (%)
	Description: Overall classifier performance, 200 iterations
- Units of measurement: unitless and %
- Number of Rows: 202
- Column Names: 
	A. True Positive Rate
	B. False Positive Rate
	C. All Balanced ACC values

Figure 1I: Change in Cluster Distribution 
- Variable:
	1. Change in cluster distribution 
	Description: Change in behavioral cluster composition from baseline (BL) for each 	condition.
- Units: %
- Rows: 51
- Column Names: 
	A. Mouse ID
	B. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	C. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D. Change in Cluster Distribution (% from BL) 

Figure 1J: Time Spent per Cluster
- Variable:
	1. % of time per cluster
	Description: Average percent of time spent in each of the 48 clusters over a 10-	minute session.
- Units: %
- Number of Rows: 47
- Column Names: 
	A. Mouse ID
	B. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	C. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D-AY. Move cluster 1-48

-----------
METHODOLOGY
-----------
Experimental Setup and Data Collection
Open-field behavioral experiments were conducted in a 40×40 cm white arena with transparent acrylic sides and floor, housed inside a sound-attenuating chamber. Mice were recorded at 30–40 fps from below using a Flea3 camera (Point Grey Research). Animals were habituated to the equipment for 2–3 days prior to testing.
On the experiment day, mice were lightly anesthetized for mounting of a wireless IMU and microendoscope. Each session consisted of a 10-min baseline (BL) recording, followed by a 10-min recording 20 minutes post-injection of either L-DOPA or vehicle (VEH). Head-mounted devices remained in place throughout.
Wireless IMUs (Champalimaud Hardware Platform, WEAR system) included 9-axis sensors (accelerometer, gyroscope, magnetometer; magnetometer unused). Devices (~1.8 g) sampled at 200 Hz and were synchronized with video and calcium imaging using Bonsai v2.4 (https://bonsai-rx.org/) and custom MATLAB scripts. Hardware sync was managed via a HARP base station.

Data Processing
Behavioral features (e.g., body acceleration, gravitational acceleration) were computed using custom MATLAB and Python scripts (available upon request). Unsupervised behavioral clustering was performed with the algorithm published at:
https://zenodo.org/records/13990593

----------------------------------------
DATA SPECIFIC INFORMATION FOR FIGURE 2
----------------------------------------
Figure title: Behavioral clustering captures specific dyskinesias and other pathological behaviors.
Date of Creation: 2025-05-27
Panels Covered: Figure 2B, 2D, 2E, 2F, 2G and 2H.
Figure 2B: Correlations between Annotations and Clusters
- Variable:
	1. Correlation annotation x cluster 
	Description: Example cluster for one mouse that significantly correlated with the 	axial+limb annotation but not with axial alone or limb alone. Correlation between 	cluster occurrences and dyskinesia annotations (axial+limb, axial alone, limb 		alone), based on bootstrap analysis vs. shuffled data (5000 shuffled iterations).
- Units: unitless (correlation measure)	
- Number of Rows: 5002
- Column Names: 
	A. Shuffled or Original Values: Original or Shuffled
	B. r Axial+limb
	C. r Axial Alone
	D. r Limb alone 

Figure 2D: Behavioral Features Quantification
- Variables (4):
	1. Body Acceleration (BA) 
	Description: Total body acceleration was calculated as the square root of the sum 	of squared filtered accelerometer components.
	Units: g
	2. Gravitational Acceleration (GA) 
	Description: Low-pass filtered gravitational component, used for postural changes 	(i.e., vertical head position) in the open field during resting, locomotion and 	rearing.
	Units: g
	3. Head angle (θ head)
	Description: Head angular velocity derived from IMU gyroscope, cumulative over 		head angle from the beginning until the end of a head rotation for all rotations 	with a minimum of ±10 deg/sec (Fugre S2A).
	Units: degrees
	4. Axial angle (θ axial)
	Description: Calculated using DeepLabCut tracking of posture (nose, hindlimbs, 		tail).
	Units: degrees
- Number of Rows: 66
- Column Names: 
	A. Mouse ID
	B. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	C. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D. Cluster group: Axial+limb, Axial alone, Limb alone, Path rot.
	E. BA
	F. GA
	G. θ head
	H. θ axial

Figure 2E: Group-Level Time in Cluster Groups
- Variable:
	1. Time per cluster group (%)
	Description: Average percentage of time mice spent in each behavioral cluster 		group during 10-minute sessions, across the four experimental conditions. The 		values represent group-level means; individual mouse data are not included, and 	no statistical analyses were conducted on these averages. For statistical 		comparisons (e.g., INT-LD vs. LES-LD), refer to Figure 2F.
- Units: %	
- Number of Rows: 21
- Column Names: 
	A. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	B. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	C. Cluster group: Axial+limb, Axial alone, Limb alone, Path rot.
	D. Time per Cluster Group (%)

Figure 2F: Time per Cluster Group Int-LD vs. Les-LD
- Variable:
	1. Time per cluster group (%)
	Description: Percentage of time each mouse spent in each cluster group per 10-		minute session.
- Units: %
- Number of Rows: 115
- Column Names: 
	A. Mouse ID
	B. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	C. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D. Cluster Group: Axial+limb, Axial alone, Limb alone, Path rot.
	E. Timer per Cluster Group (%)

Figure 2G: Classifier Accuracy
- Variable:
	1. Accuracy 
	Description: Accuracy of SVM classifier in predicting annotation type based on 		dyskinesia cluster group features, using leave-one-mouse-out cross-validation and 	a fraction of 0.8 and 0.2 for training and testing, respectively. Importantly, 		data for training and testing was balanced by subsampling the data repeatedly and 	calculating the average accuracy (n = 5000).
- Units: Unitless (proportion between 0 and 1)
- Number of Rows: 5
- Column Names: 
	A. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	B. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	D. Annotation type: Axial+limb or Axial alone annotation
	D. Cluster Group: Axial+limb or Axial alone
	E. Accuracy

Figure 2H: Correlation Between Cluster and AIMs scores
- Variable:
	1. Correlation AIMs scores x cluster 
	Description: Correlation between total AIM scores (axial, limb) and time spent in 	each behavioral cluster during the 10-minutes recording.
- Units: unitless (Pearson or Spearman correlation coefficient)
- Number of Rows: 41
- Column Names: 
	A. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	B. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	C. AIMs Score Type: Axial or Limb AIMs
	D. Cluster Group: Axial+limb or Axial alone
	E. Cluster #: 1-9 for Axial+limb and 1-11 for Axial alone
	F. Spearman rho
	G. p-value

-----------
METHODOLOGY
-----------
Data Collection
Same as described for Figure 1.

Data Processing
Behavioral feature quantification, cluster time quantification, classifier accuracy, and correlation with AIM scores were performed using custom Python scripts, available upon request from the lead contact.

----------------------------------------
DATA SPECIFIC INFORMATION FOR FIGURE 3
----------------------------------------
Figure title: L-DOPA oppositely modulates average SPN activity in 6-OHDA lesioned mice. 
Date of Creation: 2025-05-27
Panels Covered: Figure 3C, 3D, 3E, 3F, 3G, 3H and 3I.

Figure 3C&D: SPN Event Rate During Movement and Rest
- Variables (2):
	1. Rate move 
	Description: Event rate per neuron during moving periods (based on deconvolved 		Craw signals).
	Units: events/s
	2. Rate rest
	Description: Event rate per neuron during resting periods (defined by immobility 	criteria; BA < 0.05 g for ≥ 500 ms).
	Units: events/s

- Number of Rows: 51
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Rate Move (events/s)
	G. Rate Rest (events/s)

Figure 3E&F: Number of Active Neurons
- Variables:
	1. Active neurons
	Description: Count of active neurons (CNMF-E extracted) exceeding a threshold of 	0.05 events/s.
- Units: absolute number
- Number of Rows: 51
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Active Neurons (absolute number)

Figure 3G-I: Spatiotemporal Cross-correlation
- Variables:
	1. Cross-correlation
	Description: Pearson correlation between Craw traces of neuron pairs, binned by 	spatial distance.
- Units: unitless (Pearson’s r)
- Number of Rows: 27
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F - N. CC Distance Bins: 20, 31, 47, 73, 111, 171, 263, 403, 619 (µm)

-----------
METHODOLOGY
-----------
Data Collection
Same experimental setup as Figure 1. In addition, manual annotations of axial and limb AIMs were performed using Python Video Annotator on 10-min video sessions (1-2 sessions per mouse). Orofacial AIMs were not annotated due to bottom-view camera limitations. Events were categorized into axial alone, limb alone, and axial+limb, with limb alone being rare.

Data processing 
Calcium data were preprocessed using Mosaic and Inscopix software, followed by custom MATLAB scripts for motion correction and binning. Neuronal activity was extracted using CNMF-E. Matching of neuron identities between BL and treatment was done via footprint correlation and manual validation. Event rates were computed using MLspike, accounting for calcium decay and drift. Resting state detection was based on BA thresholds (< 0.05 g). Spatiotemporal cross-correlations were computed from Pearson coefficients and binned by inter-neuron distance (15-750 µm).

----------------------------------------
DATA SPECIFIC INFORMATION FOR FIGURE 4
----------------------------------------
Figure title: Specific D1-SPNs and D2-SPNs ensembles are associated to dyskinesia and pathological rotations. 
Date of Creation: 2025-05-27
Panels Covered: Figure 4A, 4B, 4C, 4D, 4E and 4F.

Figure 4A: SPN Event Rates per Behavioral Cluster Group
- Variables:
	1. Rate 
	Description: Average event rate of D1- and D2-SPNs per cluster group in Les-LD 		mice, and their ratio (right plot in Figure 4A).
- Units: events/s
- Number of Rows: 53
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G. Rate (events/s)

Figure 4B: Decoding Dyskinesia and Pathological Rotations from SPN Activity 
- Variables (2):
	1. Accuracy of original data
	Description: Accuracy of the SVM classifier to classify (1) dyskinesia vs. Other 	N and (2) Path rot vs. Other N, using deconvolved activity of original data.
	Units: unitless (proportion between 0 and 1)
	2. Accuracy of shuffle data
	Description: as for accuracy of original data.
	Units: unitless (proportion between 0 and 1)

- Number of Rows: 27
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Decoding: Dykinesia vs. Other N; Path rot vs. Other N
	G. Original: Original data accuracy
	H. Shuffle: Shuffled data accuracy

Figure 4C: Percentage of Behavior-Related Neurons
- Variable:
	1. % behavior-related neurons
	Description: Percentage of SPNs significantly modulated during specific cluster 	groups. The percentage was taken among all SPNs detected during both BL and LD 		sessions.	
- Units: %
- Number of Rows: 53
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G. Behavior-related Neurons (% of all SPNs during BL+LD)

Figure 4D: Overlap Between Behavior-Related SPNs (Original)
- Variable:
	1. % of overlap (original)
	Description: Percentage of overlap between behavior-related SPNs associated with 	different cluster groups (e.g., axial+limb vs. other N).
	Calculation:
	Overlap = 100 × (#neurons associated with both groups / #neurons associated with 	either group) 
	• An overlap of 0% indicates that no neurons are shared between groups.
	• An overlap of 100% indicates that all neurons associated with one group are 		also associated with the other.
- Units: %
- Number of Rows: 79
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion: This column shows if the mouse was lesioned with 6-OHDA (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group comparison: Axial+limb vs. Axial Alone, Axial+limb vs. Path rot, 	Axial+limb vs. Other N, Axial alone vs. Path rot, Axial alone vs. Other N, Path 	rot vs. Other N
	G. % Overlap (original)

Figure 4E: Overlap Between Behavior-Related SPNs (Shuffled-Corrected)
- Variable:
	1. % of overlap (shuffled-corrected)
	Description: Shuffled-corrected percentage of overlap between behavior-related 		SPNs from different cluster groups. Correction was performed by subtracting the 	mean overlap from 5000 iterations of shuffled data, in which the identity of the 	behavior-related neurons was randomized independently per cluster group, 		preserving average activity.
- Units: %
- Number of Rows: 79
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group comparison: Axial+limb vs. Axial Alone, Axial+limb vs. Path rot, 	Axial+limb vs. Other N, Axial alone vs. Path rot, Axial alone vs. Other N, Path 	rot vs. Other N
	G. % Overlap (shuffled-corrected)

Figure 4F: Temporal Dynamics of SPN Activity Aligned to Behavior Onset
- Variable:
	1. Averaged z-scored ΔF/F
	Description: D1- and D2-SPNs Z-scored ΔF/F aligned to the onset of axial+limb, 		axial alone, and path rot behaviors. 
- Units: ΔF/F (z-score), Time (seconds)
- Number of Rows: 40
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G – AU. Time frames from -1 to +1 s, sampled at 20 Hz (i.e., 41 frames from -20 	to +20)

-----------
METHODOLOGY
-----------
Data collection
As described in Figure 1.

Data processing
Decoding Accuracy (Panel B): A linear SVM classifier was trained on deconvolved SPN activity averaged over behavioral segments to classify (1) dyskinesia vs. other N and (2) path rot vs. other N. Accuracy was computed over 5000 iterations with balanced class sizes. Shuffled controls were generated by randomly permuting neuron identities.
Behavior-Related Neuron Overlap (Panels D, E): The percentage overlap of SPNs active across pairs of behavioral clusters was calculated as 100 × (neurons related to both clusters) / (neurons related to either cluster). Shuffle-corrected overlap was obtained by subtracting the average overlap from 5000 iterations of neuron identity shuffling within each behavioral group.
Temporal Dynamics (Panel F): Calcium signals (sampled at 20 Hz) were aligned to behavioral event onset and plotted over a −1 to +1 second window (columns G–AU in the Excel file represent frames −20 to +20).

----------------------------------------
DATA SPECIFIC INFORMATION FOR FIGURE 5
----------------------------------------
Figure title: Activity features in SPN ensembles tuned to specific dyskinesia types. 
Date of Creation: 2025-05-27
Panels Covered: Figure 5A, 5B, 5C and 5D.

Figure 5A: Rate of Behavior-Related Neurons
- Variable:
	1. Rate 
	Description: Average event rates of behavior-related D1- and D2-SPNs during each 	behavioral cluster group.
- Units: events/s	
- Number of Rows: 209
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G. Behavior-related Neurons: Axial+limb neurons, Axial alone neurons, Path rot 		neurons, Other N neurons
	H. Rate (events/s)
	I. Outlier. TRUE or FALSE

Figure 5B: Activity in Baseline-Other N Clusters
- Variable:
	1.  Active neurons in BL-other N
	Description: Percentage of the axial+limb, alxial alone and path rot-SPNs 		associated with at least one other N behavioral cluster during baseline (named 		BL-other N).
- Units: %
- Number of Rows: 40
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G. Behavior-related Neurons: Axial+limb neurons, Axial alone neurons, Path rot 		neurons, Other N neurons
	H. Active Neurons in BL-Other N (%)

Figure 5C: Rate in Baseline-Other N
- Variable:
	1.  Rate
	Description: Average event rates of behavior-related D1- and D2-SPNs during Other 	N cluster group at baseline (BL-other N).
- Units: Events/s
- Number of Rows: 40
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Cluster Group: Axial+limb, Axial alone, Path rot, Other N
	G. Behavior-related Neurons: Axial+limb neurons, Axial alone neurons, Path rot 		neurons, Other N neurons
	H. Rate (events/s)

Figure 5D: EMD Similarity
- Variable:
	1.  EMD similarity
	Description: Earth Mover’s Distance (EMD) similarity between dyskinesia clusters 	and BL-other N clusters, comparing positively modulated neurons (“pos mod”) vs. 	non-modulated (“other”).
	Units of measurement: Events/s
	
- Number of Rows: 79
- Column Names: 
	A. Mouse ID
	B. Genotype: D1-Cre or A2a-Cre
	C. SPN type: D1 (D1-SPN) or D2 (D2-SPN)
	D. Lesion. This column shows if the mouse was lesioned with 6-OHDA  (LES) or not 	(INT from intact)
	E. Treatment: Mice received vehicle (VEH) or L-DOPA (LD)
	F. Behavior-related Neurons: Axial+limb neurons, Axial alone neurons, Path rot 		neurons, Other N neurons
	G. Cluster type: “pos mod” or “other”
	H. EMD sim
	I. Outlier. TRUE or FALSE

-----------
METHODOLOGY
-----------

Data collection
Same as Figure 1.

Data Processing
Event Rates (Panels A, C): Computed using deconvolved Craw signals from SPNs classified as behavior-related based on activity during specific cluster groups.
BL-Other N Overlap (Panel B): For each ensemble (axial+limb, axial alone, path rot), we calculated the percentage of neurons that were also active in baseline other N clusters.
EMD Similarity (Panel D): Earth Mover’s Distance was computed between behavioral signatures of dyskinesia clusters (Les-LD) and BL-other N clusters. The comparison was made between “pos mod” clusters (with increased activity during dyskinesia) and all other BL clusters.


-----------------------
DATA ACCESS AND SHARING
-----------------------
Publications based on this dataset: Alcacer et al., Cell Rep 2025
Recommended citation for this dataset: Dataset for Alcacer et al., Cell Reports 2025
License information: CC-BY 4.0