Skip to content

Getting Started with SynthBioData

Welcome to SynthBioData! This guide will help you get up and running with synthetic biological data generation for drug discovery and machine learning applications.

Important Notice

This package generates synthetic data for testing and educational purposes only.

The data produced does not represent real biological or chemical measurements and should not be used for clinical, regulatory, or production applications.

What is SynthBioData?

SynthBioData is a Python package that generates realistic synthetic drug discovery data, including:

  • Molecular Descriptors: Molecular weight, LogP, TPSA, hydrogen bond donors/acceptors, and more
  • ADME Data: Absorption, Distribution, Metabolism, and Excretion properties
  • Target Families: Support for GPCR, Kinase, Protease, and other protein families
  • Chemical Fingerprints: Binary chemical fingerprints as features

Quick Installation

Install SynthBioData using your preferred package manager:

uv pip install synthbiodata

pip install synthbiodata

conda install -c conda-forge synthbiodata

How to use synthbiodata

1. Basic Usage

Start with the simplest approach by generating data with the default settings:

from synthbiodata import generate_sample_data

# Generate molecular descriptor data
df = generate_sample_data(data_type="molecular-descriptors")
print(f"Generated {len(df)} samples with {len(df.columns)} features")

# Generate ADME data
df_adme = generate_sample_data(data_type="adme")
print(f"Generated {len(df_adme)} samples with {len(df_adme.columns)} features")

2. Custom Configuration

For more control over data generation, provide a custom configuration. In the example below, custom_config has parameters like n_samples, postive_ratio and imbalanced to generate an imabalance dataset with 1000 samples:

from synthbiodata import create_config, generate_sample_data

# Create a custom configuration
custom_config = create_config(
    data_type="molecular-descriptors",
    n_samples=1000,
    positive_ratio=0.1,
    imbalanced=True,
)

# Generate data with your custom configuration
df = generate_sample_data(config=custom_config)
print(f"Total samples: {len(df)}")
print(f"Features: {len(df.columns) - 1}")  # Exclude target column
print(f"Positive ratio: {df['binds_target'].mean():.1%}")

3. Reproducible datasets

Ensure reproducible data generation by setting the paramater random_state:

# Generate reproducible data
df1 = generate_sample_data(
    data_type="molecular-descriptors",
    random_state=321
)

# Same seed = identical results
df2 = generate_sample_data(
    data_type="molecular-descriptors", 
    random_state=321
)

assert (df1 == df2).all().all()  # True

Data Types Overview

Molecular Descriptors

Generate synthetic molecular data with features like:

  • Physical Properties: Molecular weight, LogP, TPSA
  • Structural Features: Hydrogen bond donors/acceptors, rotatable bonds
  • Chemical Properties: Aromatic rings, chemical fingerprints
  • Target Information: Protein families (GPCR, Kinase, Protease, etc.)

ADME Data

Generate ADME (Absorption, Distribution, Metabolism, Excretion) data with:

  • Absorption: Bioavailability percentages and absorption rates
  • Distribution: Plasma protein binding and volume of distribution
  • Metabolism: Clearance rates and half-life predictions
  • Excretion: Renal clearance and elimination parameters

Next Steps

Now that you have the basics, explore these detailed guides:

Key Features

  • Realistic Data

    Generate data that mimics real-world molecular properties and distributions

  • Multiple Target Types

    Support for various protein families and target types

  • Configurable Parameters

    Customize data generation to match your specific needs

  • High Performance

    Built on Polars for fast data manipulation and processing

  • Type Safe

    Full type hints and Pydantic validation for robust configuration

  • :material-reproducible: Reproducible

    Deterministic data generation with random seed support

Common Use Cases

  • Machine Learning Research: Generate training data for drug discovery ML models
  • Algorithm Testing: Test and validate ML algorithms with controlled synthetic data
  • Educational Purposes: Learn about molecular properties and drug discovery concepts
  • Benchmarking: Create standardized datasets for comparing different approaches
  • Prototype Development: Quickly generate data for proof-of-concept applications

Need Help?

Ready to dive deeper? Start with the Quick Start Tutorial for a hands-on walkthrough!