CoNECTOME Hackathon

Allen Institute/Neural Dynamics

May 20, 2026

Dynamic Routing

Source: https://github.com/AllenNeuralDynamics/conect-dev-capsule

For data access outside of Code Ocean, see REMOTE.md.

Project overview

The Dynamic Routing project aims to uncover the neural mechanisms of flexible decision making. We train mice to perform a visual–auditory switching task in which the rewarded sensory modality alternates within a session, requiring mice to dynamically route sensory information to appropriate motor outputs depending on behavioral context. While mice perform this task, we record from neurons across the mouse brain to identify the neural correlates of flexible sensory-motor associations and to understand how context representations are generated, maintained, and used to guide behavior.

Recording strategy

We insert up to six Neuropixels 1.0 (NP 1.0) probes simultaneously across the mouse brain using the SHIELD implant (Bennett et al., Neuron 2024), a chronic implant system that enables repeatable, multi-probe access to dorsal cortex and underlying structures. Probes are inserted at varying angles and anterior-posterior positions to achieve broad coverage spanning cortical, thalamic, basal ganglia, and midbrain structures (among others). This approach allows us to simultaneously sample the activity of hundreds of neurons across functionally distinct brain regions while the mouse performs our switching task.

---

Diagram of SHIELD implant with multiple Neuropixels 1.0 probes.	Probe trajectories across mice and sessions showing dense coverage of left hemisphere.

---

Task

Mice perform a visual-auditory context-switching task in which the rewarded stimulus modality alternates across blocks within each session.

Stimuli and Reward Contingencies

Each trial presents one of four stimuli: a visual target (VIS+), a visual non-target (VIS−), an auditory target (AUD+), or an auditory non-target (AUD−). Trials are organized into alternating auditory-rewarded (A) and visual-rewarded (V) blocks. Within each block, only one target stimulus is rewarded (Figure below, panel a):

• Auditory context (A): licking in response to AUD+ earns a water reward; licking to VIS+ is a false alarm.
• Visual context (V): licking in response to VIS+ earns a water reward; licking to AUD+ a false alarm.

Responses to either non-target stimulus (AUD−, VIS−) are never rewarded regardless of context.

Session and Block Structure

A session lasts approximately 60 minutes and consists of six ~10-minute blocks alternating between auditory- and visual-rewarded contexts (Figure below, panel b). Context is not explicitly cued on every trial — the mouse must infer and maintain its representation of the current block from trial outcomes.

Block transitions are signaled by 5 consecutive presentations of the newly rewarded target stimulus (purple in panel b). If the mouse does not earn a contingent reward by licking during these trials, a non-contingent reward is given at the end of the response window. Following these trials, all four stimuli are presented in pseudorandom interleaved order for the remainder of the block.

Trial Structure

Each trial begins with a quiescent period (1.5 s before stimulus onset) during which licking resets the trial. The stimulus is presented for 0.5 s, followed by a response window from 0.1–1.0 s post-stimulus onset. The trial concludes with an inter-trial interval (ITI) of 3–7.5 s before the next trial begins.

a, Reward contingencies for the auditory (A) and visual (V) contexts. Only the target of the currently rewarded modality yields reward. b, Nested timeline of a session (six alternating 10-minute blocks), a block (five rewarded target cue trials followed by pseudo-randomly interleaved stimuli), and a trial (quiescent period, stimulus, response window, ITI).

Data in this capsule

This dataset comprises five Neuropixels recording sessions from five different mice, each packaged in a "data asset" in /data.

Each attached asset contains:

• an .nwb file saved as a .hdf5 file or a .zarr folder
• nwb_contents.json detailing the internal paths within each .hdf5 file, e.g.:

  [
      "/intervals/trials",
      "/processing/behavior/running_speed",
      "/units"
  ]   
  

• AIND metadata .json files

Summary

All animals successfully switched between contexts, as demonstrated by high d' values for the rewarded modality and suppressed responses to the non-rewarded modality. A total of 6,685 QC-passing units were recorded.

Session	Subject	Sex	Age	Genotype	Strain	Experiment Day
664851_2023-11-15	664851	F	P310	Pvalb-IRES-Cre/wt;Ai32	Pvalb-IRES-Cre;Ai32	Day 3
668755_2023-08-31	668755	M	P206	wt/wt	C57BL6J(NP)	Day 4
713655_2024-08-09	713655	M	P260	Sst-IRES-Cre/wt;Ai32	Sst-IRES-Cre;Ai32	Day 5
742903_2024-10-22	742903	F	P159	Vip-IRES-Cre/wt;Ai32	Vip-IRES-Cre;Ai32	Day 2
759434_2025-02-04	759434	M	P177	VGAT-ChR2-YFP/wt	VGAT-ChR2-YFP(ND)	Day 2

Three of the five mouse lines express channelrhodopsin (ChR2) in specific inhibitory interneuron classes (Pvalb, Sst, Vip, or all GABAergic via VGAT), enabling optotagging of those cell types. One mouse (668755) is wild-type.

---

Session Structure (Epochs)

Each recording session follows a standardized sequence of epochs:

1. RFMapping (~15 min) — Receptive field mapping with visual and auditory stimuli
2. OptoTagging (~3–6 min) — Optogenetic identification of genetically-defined neurons (pre-task)
3. Spontaneous (~10 min) — Spontaneous activity, no stimuli
4. SpontaneousRewards (~10 min) — Spontaneous activity with non-contingent rewards
5. DynamicRouting1 (~60 min) — Main behavioral task
6. SpontaneousRewards (~10 min) — Post-task spontaneous with rewards
7. OptoTagging (~3–6 min) — Post-task optotagging (some sessions)
8. Spontaneous (~10 min) — Post-task spontaneous (some sessions)

---

Context Blocks and Rule Switching

The session is divided into 6 blocks (indices 0–5), alternating between two rewarded-modality contexts:

• Visual context (rewarded_modality = "vis"): The mouse must lick to visual targets (vis+ → go) and withhold licking to all other stimuli (including auditory targets aud+ → no-go).
• Auditory context (rewarded_modality = "aud"): The mouse must lick to auditory targets (aud+ → go) and withhold licking to all other stimuli (including visual targets vis+ → no-go).

Some sessions begin with visual blocks first, others with auditory blocks first. Each modality context has 3 blocks per session.

---

Trial Structure

Each trial proceeds through the following phases:

1. Quiescent period (quiescent_start_time → quiescent_stop_time): The mouse must remain still (no licking) before a stimulus is presented. Violations restart the quiescent period.
2. Stimulus presentation (stim_start_time → stim_stop_time): A visual grating, auditory stimulus, or catch (blank) is presented.
3. Response window (response_window_start_time → response_window_stop_time): The mouse can lick to report detection. A lick within this window on a go trial is a hit; on a no-go trial it is a false alarm.
4. Post-response window (post_response_window_start_time → post_response_window_stop_time): Brief post-response period.
5. Reward (reward_time, if applicable): Water reward delivered on correct go responses (hits) and on some instruction/auto-reward trials.

Trial Types

Trial Type	Description
Go	Target stimulus in the currently rewarded modality; lick = hit, no lick = miss
No-go	Non-target stimulus, or target in non-rewarded modality; lick = false alarm, no lick = correct reject
Catch	No stimulus; used to measure baseline lick rate
Instruction	Auto-rewarded trials (30 per session) at block transitions to cue the new rule

Additional trial flags: is_repeat (repeated after a miss), is_opto (optogenetic stimulation applied — 0 opto trials in these task sessions), is_contingent_reward / is_noncontingent_reward.

---

Performance Summary

Overall Trial Counts

Session	Total Trials	Go	No-go	Catch	Hits	Misses	Correct Rejects	False Alarms	Total Correct	% Correct
664851	534	144	334	56	142	2	282	52	480	89.9%
668755	524	139	329	56	137	2	242	87	430	82.1%
713655	515	136	324	55	129	7	284	40	465	90.3%
742903	538	147	348	43	131	16	317	31	491	91.3%
759434	545	144	353	48	142	2	313	40	502	92.1%

All mice performed well, with overall correct rates ranging from 82–92% and hit rates consistently high (89–99%).

d-prime by Rewarded Modality (Block-Averaged)

Session	Context	Vis d'	Aud d'	Cross-Modal d'	Hit Rate	FA Rate
664851	vis	3.54	−0.74	2.53	0.99	0.15
664851	aud	0.00	2.34	3.23	0.99	0.16
668755	vis	3.65	−0.67	1.75	0.99	0.24
668755	aud	−0.83	2.64	1.95	0.98	0.30
713655	vis	3.43	−0.22	2.98	0.94	0.05
713655	aud	−0.11	1.93	2.71	0.95	0.21
742903	vis	3.15	−0.11	2.04	0.82	0.07
742903	aud	−0.01	2.88	2.80	0.96	0.10
759434	vis	3.81	−0.22	3.69	0.99	0.03
759434	aud	−0.10	2.48	2.50	0.98	0.20

Key observations:

• All mice showed strong context-dependent discrimination: high d' for the rewarded modality and near-zero or negative d' for the non-rewarded modality, demonstrating successful task switching.
• Cross-modal d' (measuring discrimination between the target of the rewarded modality vs. the target of the non-rewarded modality) was consistently positive (1.75–3.69), confirming that animals selectively responded to the correct modality.
• Hit rates were uniformly high (0.82–0.99). False alarm rates were low to moderate (0.03–0.30), varying across animals and contexts.

---

Electrophysiology: Recorded Areas (QC-Passing Units Only)

Units were filtered by is_qc_pass = true. Each session used 5–6 Neuropixels probes.

Per-Session Summary

Session	Total Units	QC-Pass Units	# Probes	# Structures (QC-pass)
664851	3,062	1,118	5	22
668755	2,878	1,184	6	28
713655	3,577	1,666	5	20
742903	4,446	1,973	6	22
759434	2,285	744	5	17
Total	16,248	6,685	—	—

QC-Passing Units by Brain Area (per session)

664851 (Pvalb-Cre;Ai32)

Structure	QC-Pass Units
AUDp	203
MOs	137
FRP	108
CA1	95
CA3	91
ORBvl	89
VISal	77
LSc	54
AUDd	52
MOB	43
ACAd	39
SSp	28
SSs	24
ACAv	23
TEa	11
VISrl	10
CA2	9
AUDv	9
ORBl	6
LSr	6
OLF	3

668755 (Wild-type C57BL6J)

Structure	QC-Pass Units
MOs	241
MOp	166
ACAv	153
SSp	96
CP	87
VISam	73
VISp	57
RSPv	52
ACAd	41
AON	33
ORBl	32
DP	18
SCiw	15
SCig	13
OLF	13
TTd	11
HPF	11
SCsg	9
SCdg	9
PPT	9
MB	8
FRP	7
RSPd	6
SCop	5
MPT	4
PAG	3
LGv	3
RSPagl	3
SCzo	2
VISpm	2
SCdw	1
NOT	1

713655 (Sst-Cre;Ai32)

Structure	QC-Pass Units
SSs	271
MOp	210
ECT	200
MOs	188
LSr	183
CP	142
TEa	117
AUDp	109
AUDv	77
SSp	38
VISli	29
LSc	28
AUDpo	23
AD	17
PERI	16
AV	11
ACAd	7

742903 (Vip-Cre;Ai32)

Structure	QC-Pass Units
MOs	437
SSp	324
ORBvl	250
ILA	172
SSs	118
CP	101
TTd	96
ORBm	79
ORBl	75
ACAd	67
GU	54
PL	53
LSr	46
ACAv	31
VISal	20
FRP	17
CA1	13
OLF	11
CA3	7
DG	2

759434 (VGAT-ChR2-YFP)

Structure	QC-Pass Units
SSp	153
MOs	147
CP	132
ILA	62
DP	59
TTd	58
CA3	50
PL	46
CA1	18
VISli	5
TEa	4
ACAd	3
OLF	3
DG	1
ccb	1
scwm	1
STR	1

Commonly Recorded Areas Across Sessions

The most frequently recorded areas with QC-passing units include:

• Frontal cortex: MOs (secondary motor), FRP (frontal pole), ACAd/ACAv (anterior cingulate), MOp (primary motor)
• Somatosensory cortex: SSp (primary), SSs (supplemental)
• Prefrontal / orbitofrontal: ORBvl, ORBl, ORBm, ILA (infralimbic), PL (prelimbic)
• Visual cortex: VISp, VISal, VISam, VISrl, VISli, VISpm
• Auditory cortex: AUDp (primary), AUDd (dorsal), AUDv (ventral), AUDpo (posterior)
• Temporal association: TEa, ECT (ectorhinal), PERI (perirhinal)
• Hippocampus: CA1, CA2, CA3, DG
• Lateral septum: LSr, LSc
• Striatum: CP (caudoputamen)
• Thalamus: AD (anterodorsal), AV (anteroventral)
• Midbrain / superior colliculus: SCig, SCiw, SCsg, SCop, SCdg, SCzo, SCdw
• Olfactory: AON, OLF, TTd, MOB, DP
• Other: GU (gustatory), RSPv/RSPd (retrosplenial), PAG, PPT, HPF, MB

---