PatchSeq-Epilepsy: Multimodal Neuronal Excitability Analysis

Overview

This repository documents my research experience analyzing neuronal excitability using patch-clamp electrophysiology and transcriptomics. I'm building computational tools to understand how gene expression patterns relate to electrical properties of neurons, with a focus on epilepsy-relevant phenotypes.

---

Research Context

Scientific Question

How do gene-expression signatures predict neuronal excitability phenotypes?

Neurons from patients with epilepsy often exhibit abnormal firing properties. However, the molecular mechanisms linking transcriptional programs to hyperexcitability remain unclear. This project uses multimodal single-cell data to:

• Identify transcriptomic features that correlate with electrophysiological traits
• Build interpretable predictive models to understand genotype-phenotype relationships
• Guide future functional validation and therapeutic target discovery

Data Source

This work leverages the Allen Institute Patch-seq Dataset, which provides simultaneous measurements of:

• Transcriptomics: Single-cell RNA-seq (gene expression across ~20,000 genes)
• Electrophysiology: Patch-clamp recordings (firing rate, rheobase, input resistance, etc.)
• Morphology & Metadata: Cell type classification, cortical layer, species

---

What This Repository Contains

Core Workflow

The analysis pipeline is structured around Snakemake, a workflow management system that ensures reproducibility:

data/raw/                   ← Raw Allen Patch-seq datasets (transcriptomics, ephys, metadata)
    ├── patchseq_transcriptomics.csv    [~5 GB]
    ├── patchseq_metadata.csv
    └── patchseq_ephys_features.csv

src/patchseq_pipeline/      ← Main Python package
    ├── data/                 (Loading & preprocessing utilities)
    ├── analysis/             (Feature selection & dimensionality reduction)
    ├── models/               (Ridge regression, elastic net models)
    └── viz/                  (Plotting & figure generation)

scripts/                    ← Standalone analysis scripts
    ├── download_data.py          (Fetch data via AllenSDK)
    ├── build_features.py         (Normalize & standardize features)
    ├── train_model.py            (Fit predictive models)
    └── generate_figures.py       (Create plots)

results/                    ← Generated outputs (after running pipeline)
    ├── figures/              (PNG plots panels)
    ├── models/               (Trained model checkpoints & metrics)
    └── logs/                 (Execution logs & debugging info)

Key Scripts

Script	Purpose	Status
download_data.py	Fetch Patch-seq data from Allen Institute	Core
build_features.py	Quality control, normalization, feature engineering	Core
train_model.py	Fit Ridge/Elastic Net models, cross-validation	Core
generate_figures.py	Produce analysis visualizations	Core
checksums.py	Data integrity verification	Utility

---

Methods (At a Glance)

1. Data Preprocessing

• Load raw gene counts and electrophysiology recordings
• Filter cells by QC metrics (library size, gene counts, mitochondrial content)
• Log-normalize gene counts: $\log(x + 1)$
• Identify and remove outliers

2. Dimensionality Reduction

• Compute variance for each gene; select top N by expression variance
• Apply PCA to reduce noise and improve model generalization
• (Optional) UMAP visualization for interactive exploration

3. Predictive Modeling

• Target Variables: Firing rate, rheobase, input resistance
• Model Choice: Ridge Regression ($\ell_2$ regularization)
• Training: 80/20 split with 5-fold cross-validation
• Metrics: R², MSE, feature importance

4. Biological Interpretation

• Extract top predictive genes (highest absolute model coefficients)
• Map genes to known epilepsy risk loci (future: pathway enrichment)
• Validate findings against literature

---

Quickstart

Prerequisites

• Python 3.9+ (recommended: 3.11)
• Mamba or Conda (for environment management)
• ~50 GB disk space (for full dataset)

---

Example Results

After running the pipeline, you'll get:

1. Feature Importance Plot – Top genes predicting firing rate
2. PCA Visualization – Transcriptome structure by electrophysiology phenotype
3. Model Performance – R² scores, residual distributions
4. Gene Lists – Ranked by predictive importance for neuroinflammatory follow-up

---

Contact & license

• MIT License — see LICENSE