biospc 0.1.0

biospc

Bio-SPC

Bio-SPC — A Python library for identifying cliff points along time/pseudotime trajectories based on single-cell RNA-seq data.

Overview

Bio-SPC leverages a 1D ResNet architecture to perform feature extraction and dimensionality reduction on single-cell gene expression matrices. By adopting a sliding-window scanning strategy across the time/pseudotime axis, it integrates three quantitative metrics, including Bhattacharyya distance, Wasserstein distance, and Isolation Forest anomaly score. These indicators are adaptively fused into a unified CPI (Cliff Point Index) curve via the entropy weight method. This framework enables the automatic detection of critical cliff points during biological processes including development, senescence, and disease progression. Corresponding key cells can be further extracted for downstream analytical workflows.

Install

pip install biospc

Quick Start

import scanpy as sc
import pandas as pd
from biospc import bio_spc_pipeline

# 1. Loading data
adata = sc.read("your_data.h5ad")
metadata = pd.read_csv("your_metadata.csv", index_col=0)
adata.obs['Pseudotime'] = metadata['Pseudotime']

# 2. Running Bio-SPC Pipeline
peak_report = bio_spc_pipeline(
    adata=adata,
    time_col='Pseudotime',
    random_seed=42,
    do_plot=True,
)

# 3. Viewing Results
print(peak_report[['Rank', 't', 'Confidence']])

Core Parameters

Parameter	Default	Description
time_col	'Pseudotime'	Name of the pseudotime column
out_dim	30	Output dimension of the ResNet
hidden_dim	128	Hidden layer dimension of the ResNet
epochs	1000	Number of training epochs for the ResNet
lr	1e-4	Learning rate
step_ratio	0.02	Step size ratio for the sliding window
safe_margin_ratio	0.02	Safety margin ratio at the edges
min_cells_ratio	0.025	Minimum cell count ratio
smooth_window	7	Window size for Savitzky–Golay smoothing
prominence	0.1	Peak prominence
distance	5	Minimum distance between peaks
extract_ratio	0.04	Half-window ratio for cell extraction
random_seed	42	Global random seed

Output

• Peak Report: A DataFrame containing Rank, time/pseudotime point t, and Confidence.
• CSV File: A list of cell barcodes corresponding to each cliff point.
• Visualization: CPI curve with multi-peak annotations (optional).

Dependencies

• Python >= 3.8
• numpy, pandas, scanpy, anndata
• torch >= 1.10
• scikit-learn, scipy, matplotlib

License

MIT