File size: 8,425 Bytes
c0d62cf
 
 
 
 
5603aed
c0d62cf
 
 
 
 
 
 
 
 
 
 
 
97158a8
 
 
 
 
 
 
 
 
 
 
 
 
 
17f5a2f
 
 
 
 
 
 
 
 
 
 
 
 
 
589ef69
17f5a2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
97158a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
---
license: cc-by-4.0
task_categories:
- feature-extraction
- time-series-forecasting
viewer: false
language:
- en
tags:
- neuroscience
- calcium-imaging
- biology
- multimodal
pretty_name: MICrONS Functional Activity Dataset (14 Sessions)
size_categories:
- 1K<n<10K
---

# MICrONS Functional Activity Dataset & Reader

This repository contains a curated portion of the MICrONS (Multi-Scale Networked Analysis of Cellular Responding Order) dataset. It consists of functional calcium imaging data from the visual cortex of mice in response to various visual stimuli (natural clips (Clip) and parametric videos (Monet2, Trippy)).  
Videos have been downsampled to match neural activity scan frequency with frame choice corresponding to the the frame appearing at least 66ms before scan time.  
The data is organized into a highly efficient, indexed HDF5 format, allowing for rapid cross-session analysis based on either stimulus identity or brain anatomy.

## 📊 Dataset Overview

- Sessions: 14 sessions of registered neural activity.
- Stimuli: Three categories of videos (Clip, Monet2, Trippy) identified by unique condition hashes.
- Neural Data: Calcium traces (responses) from thousands of neurons across multiple visual areas (V1, AL, LM, RL).
- Behavioral Data: Synchronized treadmill speed and pupil radius.
- Eye Tracking: Pupil center coordinates (x, y) for gaze analysis.

```text
root/
├── 📂 BRAIN_AREAS/                              # Anatomical Index
│   └── 📂 <area_name>/                          # e.g., V1, AL, LM, RL
│       └── 🔗 <session_id>                      -> /sessions/<session_id>

├── 📂 SESSIONS/                                 # Primary Data Storage
│   └── 📂 <session_id>/                         # e.g., 4_7, 5_6
│       ├── 📂 META/                             # Session-wide Metadata
│       │   ├── 📂 AREA_INDICES/                 # Pre-calculated neuron masks
│       │   │   └── 📄 <area_name>               [Dataset: (N_area_neurons,)]
│       │   ├── 📄 brain_areas                   [Dataset: (N_total_neurons,)]
│       │   ├── 📄 coordinates                   [Dataset: (N_total_neurons, 3)]
│       │   ├── 📄 unit_ids                      [Dataset: (N_total_neurons,)]
│       │   ├── 📄 condition_hashes              [Dataset: (N_trials,)]
│       │   └── (Attr) fps                       [Float: Sampling rate]
│       └── 📂 TRIALS/                           # Individual trial folders
│           └── 📂 <trial_idx>/                  # Chronological trial index
│               ├── 📄 responses                 [Dataset: (N_neurons, F_trial)]
│               ├── 📄 behavior                  [Dataset: (2, F_trial)]
│               ├── 📄 pupil_center              [Dataset: (2, F_trial)]
│               └── (Attr) condition_hash        [String: Reference to video]

├── 📂 TYPES/                                    # Stimulus Category Index
│   └── 📂 <stim_type>/                          # e.g., Clip, Monet2, Trippy
│       └── 🔗 <encoded_hash>                    -> /videos/<encoded_hash>

└── 📂 VIDEOS/                                   # Stimulus Library (Saved once)
    └── 📂 <encoded_hash>/                       # Encoded version of condition_hash
        ├── 📄 clip                              [Dataset: (Frames, H, W)]
        ├── 📂 INSTANCES/                        # Reverse-index to trials
        │   └── 🔗 <session_id>_tr<trial_idx>    -> /sessions/<session_id>/trials/<trial_idx>
        └── (Attr) original_hash                 [String: The raw hash]
        └── (Attr) type                          [String: Stimulus type]
```

## ⚙️ Setup & Installation

There are two ways of accessing the contents of this repo. 

### 1. Clone the Repository
Since the dataset is stored as a large HDF5 file (.h5), you must have Git LFS installed.

```bash
# Install git-lfs if you haven't already
git lfs install

# Clone the repository
git clone https://huggingface.co/datasets/NeuroBLab/MICrONS
cd microns-functional

# Install required packages

pip install - r requirements.txt
```
### 2. Programmatic Access (Python)

If you don't want to clone the full repository, you can download the reader and the data file directly into your Python script using the huggingface_hub library.

**1. Install the library**

```bash
pip install huggingface_hub h5py numpy
```

**2. Download and Run**

```python
import sys
import importlib.util
from huggingface_hub import hf_hub_download

# 1. Define Repository Info
REPO_ID = "NeuroBLab/MICrONS"
DATA_FILENAME = "microns.h5"
READER_FILENAME = "reader.py"

print("Downloading files from Hugging Face...")

# 2. Download the Reader script
reader_path = hf_hub_download(repo_id=REPO_ID, filename=READER_FILENAME)

# 3. Download the HDF5 Data file (this handles Git LFS automatically)
data_path = hf_hub_download(repo_id=REPO_ID, filename=DATA_FILENAME)

# 4. Dynamically import the MicronsReader class from the downloaded file
spec = importlib.util.spec_from_file_location("reader", reader_path)
reader_module = importlib.util.module_from_spec(spec)
sys.modules["reader"] = reader_module
spec.loader.exec_module(reader_module)

from reader import MicronsReader

# 5. Use the reader
with MicronsReader(data_path) as reader:
    print("File downloaded and reader initialized!")
    reader.print_structure(max_items=1)
```

## 🛠️ Reader API Demo

The MicronsReader class is designed to handle the hierarchical structure of the HDF5 file transparently, including internal hash encoding and SoftLink navigation.

### 1. Initialize the Reader

The best way to use the reader is via a context manager to ensure the HDF5 file handle is closed properly.

```python
from reader import MicronsReader

path = "microns.h5"

with MicronsReader(path) as reader:
    # Your analysis code here
    pass
```

### 2. Overview of Dataset Structure

To see the internal organization of the file without loading the actual data into RAM:

```python
with MicronsReader(path) as reader:
    reader.print_structure(max_items=3)
```

### 3. Exploring Stimuli and Sessions

You can query the database by session, stimulus type, or brain area.

```python
with MicronsReader(path) as reader:
    # List available stimulus types
    types = reader.get_video_types() # ['Clip', 'Monet2', 'Trippy']

    # Get all unique hashes for a specific type
    monet_hashes = reader.get_hashes_by_type('Monet2')

    # Find which videos were shown in a specific session
    session_hashes = reader.get_hashes_by_session('4_7', return_unique=True)

    # Check which brain areas are recorded in a session
    areas = reader.get_available_brain_areas('4_7') # ['V1', 'AL', 'LM', 'RL']
```

### 4. Loading Full Data (Stimulus + Responses)

The get_full_data_by_hash method is the most powerful tool in the library. It aggregates the video pixels and every recorded neural/behavioral repeat across all 14 sessions.

```python
target_hash = "0JcYLY6eaQxNgD0AqyHf"

with MicronsReader(path) as reader:
    # Load all data for this video, filtering for V1 neurons only
    data = reader.get_full_data_by_hash(target_hash, brain_area='V1')

    if data:
        print(f"Video Shape: {data['clip'].shape}") # (Frames, H, W)
        
        for trial in data['trials']:
            print(f"Session: {trial['session']}")
            print(f"Neural Responses: {trial['responses'].shape}") # (Neurons, Frames)
            print(f"Running Speed: {trial['behavior'][0, :]}")
```

## 📂 Internal HDF5 Structure

The database is structured to minimize redundancy by storing the video "Clip" once and linking it to multiple "Trials" across sessions.
- `/videos/`: Contains the raw video arrays and links to their session instances.
- `/sessions/`: The "source of truth" for neural activity, organized by session ID and trial index.
- `/types/`: An index group for fast lookup of videos by category (Clip, Monet2, etc.).
- `/brain_areas/`: An index group linking brain regions (V1, LM...) to the sessions where they were recorded.

## 📝 Citation

If you use this dataset or reader in your research, please cite the original MICrONS Phase 3 release and this repository.