---

name: polars-bio description: A high-performance skill for processing genomic data (VCF, FASTA, BED) using polars-bio. Features streaming VCF processing, interval joins, FASTA analysis, and variant annotation. license: Apache-2.0

Polars-Bio Skill

Overview

This skill leverages polars-bio and polars to provide a robust toolkit for large-scale genomic data analysis. It is designed to handle datasets larger than memory using Polars' streaming engine and lazy evaluation.

Capabilities

1. VCF Processing

• Lazy Loading: Use pb.scan_vcf() to define execution plans without loading data into memory immediately.
• Streaming Conversion: Efficiently convert VCFs to Parquet (sink_parquet) for faster downstream queries.
• Filtering & Cleaning: Parse INFO fields and filter variants using Polars expressions.

2. Genomic Interval Operations

• Overlap Joins: High-performance interval joins (finding variants within genomic regions like Cytobands or Genes) using pb.overlap().

3. Sequence Analysis

• FASTA Processing: Lazy reading of FASTA files (pb.scan_fasta) to calculate sequence metrics (e.g., GC content, sequence length).

4. Variant Annotation

• Database Integration: Annotate VCF variants with external datasets (e.g., gnomAD, dbSNP) using efficient point-joins on chrom, start, ref, alt.

Quick Start

Basic VCF Loading & Interval Join

import polars as pl
import polars_bio as pb

# 1. Lazy Load VCF
vcf_lf = pb.scan_vcf("data/clinvar.vcf.gz")

# 2. Load Regions (e.g., BED file)
regions_lf = pl.scan_csv("data/regions.bed", separator="\t", has_header=False, 
                         new_columns=["chrom", "start", "end", "name"])

# 3. Perform Interval Overlap
# Finds variants in VCF that overlap with regions
joined_lf = pb.overlap(vcf_lf, regions_lf)

# 4. Execute (Streaming)
result = joined_lf.collect(streaming=True)
print(result)

Convert VCF to Parquet

# Efficiently convert large VCF to Parquet without high memory usage
pb.scan_vcf("input.vcf").sink_parquet("output.parquet")

FASTA GC Content Analysis

lf = pb.scan_fasta("genome.fa")
lf = lf.with_columns(
    pl.col("sequence").str.count_matches("G|C").alias("gc_count"),
    pl.col("sequence").str.len_chars().alias("len")
)
print(lf.select(pl.col("gc_count").sum() / pl.col("len").sum()).collect())

Benchmarks

The skill includes a benchmark comparing Eager vs. Streaming execution. Streaming mode has been shown to reduce memory usage by ~33% on standard workloads. See ANALYSIS_REPORT.md for details.

Dependencies

• polars
• polars-bio
• pyarrow (for Parquet)