SLSDeconv 0.1.1

Fast Spatial Transcriptomics Deconvolution

SLS - Spatial Deconvolution Package

A Python package for spatial transcriptomics cell type mapping and cell mapping using SVD ridge regression.

Installation

pip install SLSDeconv

Two Core Functions

SLS provides two main deconvolution functions:

1. Cell-Type Mapping

sd.memory_efficient_svd_ridge_regression(X, Y, celltype_labels, lambda_reg)

Maps spatial spots → cell type proportions

2. Cell Mapping

sd.memory_efficient_svd_ridge_regression_cellmap(X, Y, lambda_reg)

Maps spatial spots → individual cells (for temporal analysis)

Quick Example

import SLS as sd
import scanpy as sc
import numpy as np

# Load data
X, Y, celltype_labels = sd.load_data('FinalAnnotation', st_dir='spatial.h5ad', sc_dir='sc.h5ad')

# Cell-type mapping
lambda_opt, _, _ = sd.select_optimal_lambda(X, Y)
M_hat, celltypes = sd.memory_efficient_svd_ridge_regression(X, Y, celltype_labels, lambda_opt)

# Visualize
adata = sc.read_h5ad('spatial.h5ad')
sd.add_deconvolution_to_adata(adata, celltypes, M_hat, 'deconv')
sd.get_maximum_annotation(adata, 'deconv', 'dominant_celltype')
sd.plot_spatial_deconvolution(adata, 'dominant_celltype', palette=sd.REFERENCE_COLORS)

Complete Workflows

Cell-Type Mapping Workflow

# 1. Load data
X, Y, labels = sd.load_data('FinalAnnotation', st_dir='...', sc_dir='...')

# 2. Select optimal lambda
lambda_opt, _, _ = sd.select_optimal_lambda(X, Y)  # or sd.select_optimal_lambda_calsurrogate

# 3. Deconvolve
M_hat, celltypes = sd.memory_efficient_svd_ridge_regression(X, Y, labels, lambda_opt)

# 4. Visualize
sd.add_deconvolution_to_adata(adata, celltypes, M_hat, 'deconv')
sd.get_maximum_annotation(adata, 'deconv', 'celltype_result')

Cell Mapping Workflow (Temporal Analysis)

# 1. Filter to cell types of interest
X_filtered, Y_filtered, time_labels = sd.load_data('palantir_pseudotime', adata=adata_sub, HHA=temporal_data)

# 2. Select lambda (use mtruncation method for cell mapping)
lambda_opt, _, _ = sd.select_optimal_lambda_mtruncation(X_filtered, Y_filtered)

# 3. Cell mapping
M_temporal = sd.memory_efficient_svd_ridge_regression_cellmap(X_filtered, Y_filtered, lambda_opt)

# 4. Visualize temporal mapping
sd.add_deconvolution_to_adata(adata_sub, time_labels, M_temporal, 'temporal')
sd.get_maximum_annotation(adata_sub, 'temporal', 'pseudotime', continuous=True, feature_values=time_labels)

Key Functions

Function	Purpose
load_data()	Load spatial and single-cell data
select_optimal_lambda()	Lambda selection for cell-type mapping
select_optimal_lambda_mtruncation()	Lambda selection for cell mapping
memory_efficient_svd_ridge_regression()	Core Function 1: Cell-type mapping
memory_efficient_svd_ridge_regression_cellmap()	Core Function 2: Cell mapping
add_deconvolution_to_adata()	Add results to spatial data
get_maximum_annotation()	Get dominant cell types/values

Requirements

• Python ≥ 3.11
• scanpy, numpy, pandas, scipy, matplotlib

Usage Notes

• Cell-type mapping: Use select_optimal_lambda() or select_optimal_lambda_calsurrogate()
• Cell mapping: Use select_optimal_lambda_mtruncation() only
• Large datasets: Add output_dir and chunk_size parameters for memory efficiency
• Visualization: Built-in functions with sd.REFERENCE_COLORS palette

Example Files

See example.py for a complete reproduction of published analysis workflows.