Usage

OncoMark provides a Python API for two core functionalities:

  1. Stem Cell Type Classification – Classifies transcriptomic samples into one of three categories:

    • Pluripotent
    • Multipotent
    • Unipotent

  2. Transcriptomic Deconvolution – Estimates and quantifies the relative proportions of different cell types within bulk RNA-seq samples.

This page outlines the steps for using both modules with example data and scripts.

📁 Example Data

Example input files for testing can be found in the test directory of the GitHub repository:

  • classifier_test.csv – For cell-type classification
  • PBMC_PSEUDOBULK.csv – Bulk RNA-seq input data
  • signature_matrix_pbmc.csv – Signature matrix for deconvolution
  • cell_fractions_PBMC.csv – True cell-type frequencies
  • PBMC_PSEUDOBULK_INDEPENDENT.csv – Independent data on which trained model will be applied.

🛠️ Installation

Install the ACSCeND package:

pip install ACSCeND
````

---

## 📦 Step 1: Import Required Modules

Start by importing the required classes from the package. Make sure your transcriptomics data is loaded into a `pandas.DataFrame`.

```python
import pandas as pd
from ACSCeND import Predictor, Deconvoluter

🔬 Step 2a: Stem Cell Type Classification

The Predictor class allows you to classify transcriptomic samples into stem cell types.

⚠️ Note: The classifier handles all necessary preprocessing internally. You may use either raw counts or normalized expression data.

❗ Do not use the raw model file from GitHub directly — the Python API is required to ensure preprocessing and normalization are done correctly.

✅ Example Usage

# Load sample input data
df_pred = pd.read_csv('classifier_test.csv', index_col=0)

# Initialize classifier
predictor = Predictor()

# Predict stem cell type (returns class labels)
predictions = predictor(df_pred, prob=False)
print(predictions)

# Predict class probabilities
probabilities = predictor(df_pred, prob=True)
print(probabilities)

🧬 Step 2b: Bulk RNA-seq Deconvolution

The Deconvoluter class can be used to infer cell-type compositions from bulk RNA-seq samples.

✅ Required Inputs

  • data – Bulk RNA-seq gene expression matrix (samples × genes) [For Training]
  • sig – Signature gene expression matrix (cell types × genes) [For Training]
  • freq – Known cell-type fractions [For Training]
  • org – Independent bulk test dataset [For Inference]
  • normalized – Whether the input data is already normalized (True/False)

✅ Example Usage

# File paths
data_path = 'PBMC_PSEUDOBULK.csv'
signature_path = 'signature_matrix_pbmc.csv'
true_frequencies = 'cell_fractions_PBMC.csv'
independent_data = 'PBMC_PSEUDOBULK_INDEPENDENT.csv'

# Run deconvolution (set normalized=False for raw counts)
org_freq = Deconvoluter(
    data_path,
    signature_path,
    true_frequencies,
    independent_data,
    normalized=False
)

# org_freq contains the estimated cell-type frequencies for each bulk sample
print(org_freq)

📎 Notes

  • Input gene expression data should be in the form of a .csv file where rows are genes and columns are samples.
  • Ensure consistent gene naming across bulk data and signature matrices.
  • If you're running on a large dataset, ensure enough memory is available for processing.

📚 References

```