BAP_Moseq

Overview

This repository contains behavioral analysis data and code for the BAP project using Motion Sequencing (MoSeq). The project investigates how different early-life environmental conditions affect mouse behavior across maternal and offspring populations.

Experimental Design

Environmental Conditions

Mice were exposed to four distinct environmental conditions:

Condition	Description
EE	Enriched Environment - Enhanced sensory stimulation, motor challenges, and social interaction
LNB	Limited Nesting and Bedding - Restricted nesting materials with minimal bedding
NGH	Normal Growth Housing - Standard laboratory housing conditions (baseline control)
SI	Social Isolation - Reduced social contact with individual housing

Populations

• Moms: Maternal mice exposed to experimental conditions
• Offsprings: Offspring mice stratified by sex (Males and Females)

Repository Structure

BAP_Moseq/
├── README.md                    # This file
├── 12-1-2025_Run/              # December 2025 analysis run
│   ├── All/                    # Combined population processing
│   ├── Moms/                   # Maternal mice processing
│   └── Offsprings/             # Offspring mice processing and analysis
│       ├── code/               # Processing and analysis notebooks
│       ├── data/               # Model outputs and processed data
│       └── figs/               # Generated figures
└── 3-27-2025/                  # March 2025 analysis run
    ├── Data/                   # Shareable data outputs
    │   ├── Fingerprint_Data/   # Raw MoSeq fingerprint data
    │   └── PC_Embeddings/      # Principal component embeddings
    ├── Moms/                   # Maternal mice analysis
    │   ├── code/               # Processing and analysis notebooks
    │   └── data/               # Processed data outputs
    └── Offsprings/             # Offspring mice analysis
        ├── code/               # Processing and analysis notebooks
        ├── data/               # Processed data outputs
        └── figs/               # Generated figures

Methodology

MoSeq Pipeline

The Motion Sequencing (MoSeq) pipeline extracts quantitative behavioral features from depth-camera recordings of mouse behavior. Key steps include:

1. Extraction: Raw depth data processed to extract pose and position information
2. Modeling: Autoregressive Hidden Markov Models identify discrete behavioral syllables
3. Analysis: Syllable usage, transition probabilities, and scalar metrics quantified

Population-Specific Models

Separate MoSeq models were trained for each population:

• Moms Model: Trained exclusively on maternal behavioral data
• Offsprings Model: Trained exclusively on offspring behavioral data

This approach optimizes feature extraction for each population's behavioral repertoire.

Feature Engineering

Five behavioral metrics are extracted per mouse:

1. dist_to_center_px: Spatial positioning relative to arena center
2. height_ave_mm: Average height and rearing behavior
3. length_mm: Body length measurements
4. velocity_2d_mm: Two-dimensional movement velocity
5. MoSeq: MoSeq-specific behavioral fingerprints

Each metric produces 99 features, yielding a 495-dimensional feature matrix per mouse.

Dimensionality Reduction

Principal Component Analysis (PCA) reduces the 495-dimensional feature space:

• Moms: 8 principal components optimal
• Male Offspring: 25 principal components optimal (75.8% balanced accuracy)
• Female Offspring: 12 principal components optimal

Analysis Runs

3-27-2025 Run

The March 2025 run contains the primary analysis with:

• Complete fingerprint data for both populations
• Optimized PC embeddings with classification results
• Comprehensive documentation of methodology

12-1-2025 Run

The December 2025 run includes:

• Updated processing notebooks
• New model outputs (model-006-4641589)
• Extended analysis with additional classifier notebooks

Data Files

Key Data Formats

File Type	Description
moseq_df.csv	Main MoSeq dataframe with session metadata
stats_df.csv	Statistical summary per session
*_syllable_counts.csv	Syllable usage by group
*_transition_matrix.csv	Syllable transition probabilities
FingerprintSummary_*.csv	Multi-index behavioral fingerprint data
*_embedding.csv	Principal component coordinates

Identifier System

Each mouse is identified by:

• uuid: Universal unique identifier for cross-study compatibility
• moseq_id: BAP group-specific identifier (session format)
• SessionName: Human-readable session identifier

Dependencies

• Python 3.x
• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn
• scipy
• statsmodels
• joblib

Usage

Loading Embeddings

import pandas as pd

# Load offspring embeddings
df_male = pd.read_csv('3-27-2025/Data/PC_Embeddings/Offsprings/Male/males_final_data.csv')
df_female = pd.read_csv('3-27-2025/Data/PC_Embeddings/Offsprings/Female/females_final_data.csv')

# Access PC coordinates by condition
ee_males = df_male[df_male['category'] == 'EE']

Loading Fingerprint Data

import pandas as pd

# Load multi-index fingerprint data
summary_df = pd.read_csv('path/to/FingerprintSummary_full.csv', 
                         index_col=[0, 1], header=[0, 1])

# Extract specific population
males_df = summary_df[summary_df.index.get_level_values('group').str.contains('Male')]

References

For detailed methodology on Motion Sequencing:

• Wiltschko AB, et al. (2015) Mapping Sub-Second Structure in Mouse Behavior. Neuron.
• Markowitz JE, et al. (2018) The Striatum Organizes 3D Behavior via Moment-to-Moment Action Selection. Cell.

Contact

For questions regarding this repository, please contact the BAP research group.