MLMarker

MLMarker is a Python package for tissue-specific proteomics prediction using machine learning, with integrated SHAP-based explainability features.

Key Features

• Dual Model Support: Binary and quantitative tissue prediction models
• SHAP-Based Predictions: Uses SHAP values for more interpretable predictions
• Feature Penalty System: Adjustable penalty for absent features using penalty_factor
• Visualization Tools: Force plots, radar charts, and custom visualizations
• Protein Analysis: Integrated tools for NSAF calculation and protein information retrieval
• Data Validation: Automatic handling of missing features

Installation

pip install mlmarker

Quick Start

import pandas as pd
from mlmarker import MLMarker

# Load your data
df = pd.read_csv("your_sample.csv")

# Initialize model (binary=False for quantitative model)
model = MLMarker(binary=False, penalty_factor=1)

# Load and validate your sample
model.load_sample(df)

# Get predictions
predictions = model.predict_top_tissues_shap(n_preds=5)

Core Features

1. Model Initialization

# Binary model
binary_model = MLMarker(binary=True)

# Quantitative model with penalty for absent features
quant_model = MLMarker(binary=False, penalty_factor=1)

2. SHAP-Based Predictions

# Get predictions with SHAP explanations
predictions = model.predict_top_tissues_shap(n_preds=5)

# Visualize SHAP force plot
model.shap_force_plot(n_preds=3)

# Generate radar chart of predictions
model.radar_chart()

3. SHAP Value Analysis

# Get raw SHAP values
shap_values = model.explainability.calculate_shap()

# Get processed SHAP values with optional penalty
shap_df = model.explainability.get_shap_values(n_preds=5)

4. Feature Handling

# Get model features
features = model.get_model_features()

# Load sample with feature validation
added_features = model.load_sample(df, output_added_features=True)

5. NSAF Calculations

# Calculate NSAF scores for proteins
nsaf_df = model.explainability.calculate_NSAF(protein_df, lengths_df)

Advanced Usage

Penalty Factor

The penalty_factor parameter controls how absent features influence predictions:

• 0: No penalty (default)
• 1: Full penalty for absent features
• Values between 0-1: Partial penalty

# Model with full penalty for absent features
model = MLMarker(penalty_factor=1)

Custom SHAP Visualization

# Visualize specific tissue
model.shap_force_plot(tissue_name="Liver")

# Visualize top N predictions
model.shap_force_plot(n_preds=3)

Additional Utilities

from mlmarker.utils import (
    get_protein_info,
    get_hpa_info,
    get_go_enrichment,
    visualise_custom_plot
)

# Get protein information
protein_info = get_protein_info("P12345")

# Get Human Protein Atlas information
hpa_info = get_hpa_info("P12345")

# Perform GO enrichment analysis
enrichment = get_go_enrichment(protein_list)

Requirements

• Python ≥ 3.8
• numpy==1.23.5
• pandas
• scikit-learn
• shap==0.42.0
• plotly
• bioservices
• gprofiler-official