sumstats-liftover 1.0.0

Fast chain-based liftover for pandas DataFrames

sumstats-liftover

Fast chain-based liftover for pandas DataFrames using UCSC chain files.

A standalone, vectorized implementation for lifting over genomic coordinates in pandas DataFrames. This library provides a fast and efficient way to convert genomic coordinates from one genome build (e.g., hg19/GRCh37) to another (e.g., hg38/GRCh38) using UCSC chain files.

Note: This module is part of GWASLab, a comprehensive Python package for processing and visualizing GWAS summary statistics.

Features

• Fast: ~1.2M rows/second throughput, 24-25x faster than UCSC liftOver
• Built-in chain files: Includes commonly used chain files (hg19↔hg38, hg18→hg19)
• Standalone: No external dependencies on UCSC tools
• Flexible: Custom column names, 0-based/1-based coordinates
• Robust: Handles chromosome normalization, special chromosomes, unmapped variants
• Accurate: 100% agreement with UCSC liftOver for standard chromosomes

Installation

pip install sumstats-liftover

Requirements: Python >= 3.8, numpy >= 1.20.0, pandas >= 1.3.0

Quick Start

import pandas as pd
from sumstats_liftover import liftover_df, get_chain_path

# Create dataframe with genomic coordinates
df = pd.DataFrame({
    'CHR': [1, 1, 2],
    'POS': [725932, 725933, 100000],  # hg19 positions
    'EA': ['G', 'A', 'C'],
    'NEA': ['A', 'G', 'T']
})

# Perform liftover using built-in chain file
result = liftover_df(
    df,
    chain_path=get_chain_path("hg19ToHg38"),
    chrom_col="CHR",
    pos_col="POS"
)

print(result[['CHR', 'POS', 'CHR_LIFT', 'POS_LIFT', 'STRAND_LIFT']])

Usage

Built-in Chain Files

The package includes commonly used chain files:

from sumstats_liftover import get_chain_path, list_chain_files

# List available chain files
list_chain_files()
# {'hg19ToHg38': 'Convert from hg19/GRCh37 to hg38/GRCh38',
#  'hg38ToHg19': 'Convert from hg38/GRCh38 to hg19/GRCh37',
#  'hg18ToHg19': 'Convert from hg18 to hg19/GRCh37'}

# Use built-in chain file
result = liftover_df(df, chain_path=get_chain_path("hg19ToHg38"))

Custom Chain Files

Use your own chain files by providing the path:

result = liftover_df(df, chain_path="/path/to/custom.chain.gz")

UCSC chain files: Download

Filtering Options

Default behavior matches UCSC liftOver (allows non-standard chromosomes, alternate contigs, inter-chromosomal mappings).

Filter problematic mappings:

# Remove all problematic mappings with one parameter
result = liftover_df(df, chain_path=chain_path, remove=True)

# Or control individually
result = liftover_df(
    df,
    chain_path=chain_path,
    remove_unmapped=True,                    # Remove unmapped variants
    remove_nonstandard_chromosomes=True,    # Filter non-standard chromosomes
    remove_alternative_chromosomes=True,     # Filter alternative contigs
    remove_different_chromosomes=True        # Filter inter-chromosomal mappings
)

Coordinate Systems

# 0-based input/output (BED format)
result = liftover_df(df, chain_path=chain_path, 
                     one_based_input=False, one_based_output=False)

# 1-based input/output (GWAS standard, default)
result = liftover_df(df, chain_path=chain_path,
                     one_based_input=True, one_based_output=True)

Custom Column Names

result = liftover_df(
    df,
    chain_path=chain_path,
    chrom_col="Chromosome",
    pos_col="BP",
    out_chrom_col="CHR_hg38",
    out_pos_col="POS_hg38"
)

Special Chromosomes

By default, special chromosomes (X, Y, M) are kept as strings. Convert to numeric:

result = liftover_df(df, chain_path=chain_path, 
                     convert_special_chromosomes=True)  # X→23, Y→24, M→25

API Reference

liftover_df()

Main function for lifting over genomic coordinates.

Parameters:

Parameter	Type	Default	Description
df	pd.DataFrame	-	DataFrame with genomic coordinates
chain_path	str	-	Path to UCSC chain file
chrom_col	str	"CHR"	Input chromosome column name
pos_col	str	"POS"	Input position column name
out_chrom_col	str	"CHR_LIFT"	Output chromosome column name
out_pos_col	str	"POS_LIFT"	Output position column name
out_strand_col	str	"STRAND_LIFT"	Output strand column name
one_based_input	bool	True	Whether input is 1-based
one_based_output	bool	True	Whether output should be 1-based
remove	bool	False	Remove all problematic mappings (convenience option)
remove_unmapped	bool	False	Remove unmapped variants
remove_nonstandard_chromosomes	bool	False	Filter non-standard chromosomes
remove_alternative_chromosomes	bool	False	Filter alternative contigs
remove_different_chromosomes	bool	False	Filter inter-chromosomal mappings
convert_special_chromosomes	bool	False	Convert X→23, Y→24, M→25
ucsc_compatible	bool	False	Explicit UCSC-compatible mode (redundant with defaults)

Returns: pd.DataFrame with lifted coordinates added as new columns.

Chain File Functions

• get_chain_path(name) - Get path to built-in chain file
• list_chain_files() - List all available built-in chain files
• get_chain_info(name) - Get information about a chain file

Performance

Benchmarks

Dataset Size	Time	Throughput	Memory
1,000 rows	~0.19s	~5,200 rows/s	< 10 MB
10,000 rows	~0.19s	~54,000 rows/s	< 20 MB
1,000,000 rows	~0.84s	~1,190,000 rows/s	~200 MB
30,000,000 rows	~24s	~1,250,000 rows/s	~2 GB

Key characteristics:

• Consistent ~1.2M rows/second throughput across all sizes
• Linear scaling with dataset size
• Memory efficient: ~60-80 KB per row

Comparison with UCSC liftOver

Tool	Throughput	Time (1M)	Time (30M)	Speed
sumstats-liftover	~1.2M rows/s	0.84s	~24s	24-25x faster
UCSC liftOver	~48.6K rows/s	20.58s	~617s	Baseline

Accuracy: 100% agreement with UCSC liftOver for standard chromosome mappings (tested on 1M variants).

How It Works

The library builds a disjoint interval cover from UCSC chain files by selecting the highest-scoring segment at each position when overlaps occur. This enables O(log n) coordinate lookup using binary search.

Algorithm:

1. Parse all alignment segments from chain file
2. Build disjoint cover: for overlaps, select highest-scoring segment
3. Create sorted index for fast binary search lookup

Testing

# Run all tests
pytest tests/ -v

# Run performance tests
pytest tests/test_performance.py -v -s

# Run accuracy tests
pytest tests/test_variant_types.py -v

See example.py for usage examples.

License

Package: MIT License (see LICENSE file)

UCSC Chain Files: Built-in chain files are proprietary to The Regents of the University of California:

• Free for Independent Researchers and Nonprofit Organizations (non-commercial use)
• Commercial use requires UCSC license
• EULA | Licensing

Users are responsible for ensuring compliance with UCSC EULA.

Citation

GWASLab (main package):

@article{he2023gwaslab,
  title = {GWASLab: a Python package for processing and visualizing GWAS summary statistics},
  author = {He, Yunye and Koido, Masaru and Shimmori, Yoichi and Kamatani, Yoichiro},
  year = {2023},
  journal = {Jxiv},
  doi = {10.51094/jxiv.370}
}

sumstats-liftover:

@software{sumstats-liftover,
  title = {sumstats-liftover: Fast chain-based liftover for pandas DataFrames},
  author = {He, Yunye},
  year = {2024},
  url = {https://github.com/yourusername/sumstats-liftover},
  note = {Module of GWASLab}
}

Links

• GWASLab - Main package
• GitHub Repository
• UCSC Genome Browser