sbmutils

A collection of SBM functions.

Installation

pip install sbmutils

Features

Preprocessing

Normalization

• quantilenorm: Performs 2D quantile normalization over columns
  • Supports both mean and median averaging methods
  • Handles missing values (NaN)
  • Input validation and error handling
• stacked_quantilenorm: Performs quantile normalization on stacked data with batch information
• referenced_quantilenorm: Normalizes data using reference quantiles
• standardize: Standardizes data by centering and scaling

Biological Normalization

• comba/combat_seqt: Batch effect correction using ComBat
• counts_to_fpkm: Converts count data to FPKM

Filtering

• parse_gtf: Parses GTF files for gene information
• get_gene_id_to_entrez_mapper: Maps gene IDs to Entrez IDs
• entrez_filtering: Filters data based on Entrez IDs
• protein_coding_filtering: Filters for protein-coding genes

Decomposition

• NMF: Non-negative Matrix Factorization implementation

Usage

import numpy as np
import pandas as pd
from sbmutils.preprocess import quantilenorm, standardize, combat
from sbmutils.decomp import NMF

# Quantile normalization example
data = pd.DataFrame([[1, 4], [2, 5], [3, 6]])
normalized_data = quantilenorm(data, average="mean")

# Batch effect correction
counts = pd.DataFrame(...)  # Your count data
batch = [1, 1, 2, 2, ...]   # Batch information
corrected_data = combat(counts, batch)

# NMF decomposition
nmf = NMF(num_components=3)
nmf.fit(data)

Requirements

• Python >= 3.6
• NumPy >= 1.19.0
• SciPy >= 1.7.0
• inmoose >= 0.1.0
• pyranges >= 0.0.100
• gtfparse >= 1.2.1
• pybiomart >= 0.1.0

How to check MATLAB compatibility

1. Intall MATLAB engine for python

cd "mablab_root/extern/engines/python"
python setup.py install

2. Start MATLAB engine

import matlab
import matlab.engine
eng = matlab.engine.start_matlab()

3. Test with MATLAB function

x_matlab = matlab.double(x_python.tolist())

result_python = function(x_python)
result_matlab = eng.function(x_matlab)

np.testing.assert_array_almost_equal(result_python, result_matlab)

License

This project is licensed under the MIT License.

import matlab
import matlab.engine

"""
test code for NMF
"""
eng = matlab.engine.start_matlab()

x = np.random.randn(1000, 200)
x.ravel()[np.random.choice(x.size, 128, replace=False)] = np.nan
w = np.random.uniform(size=[1000, 4])
h = np.random.uniform(size=[4, 400])

nmf = NMF(num_components=4, num_iter=10, nmf_iter=10)
nmf.fit(x, init_w=w, init_h=h)

coph_cor_py = nmf.correlation_coefficient
ave_C_py = nmf.consensus_matrix

x_matlab = matlab.double(x.tolist())
w_matlab = matlab.double(w.tolist())
h_matlab = matlab.double(h.tolist())
out_m = eng.aoNMF_subtyping_NaN(x_matlab, 4, 10., 10., w_matlab, h_matlab, nargout=4)

coph_cor_m = np.array(out_m[0])
ave_C_m = np.array(out_m[1])

np.testing.assert_array_almost_equal(coph_cor_py, coph_cor_m)
np.testing.assert_array_almost_equal(ave_C_py, ave_C_m)