Brent S. Pedersen

This paper explores the use of LLM-based coding agents in scientific computing, presenting case studies in the life sciences and discussing their potential and challenges.

The paper uses a four-generation family to systematically assess de novo mutations and recombination, finding high mutation rates and paternal bias, along with a detailed recombination map.

sv-channels improves SV detection precision using deep learning, outperforming GRIDSS while maintaining Manta's recall.

TRGT-denovo accurately detects de novo tandem repeat mutations in family trios, reducing false positives and improving precision compared to genotype-based methods.

vcfexpress is a high-performance tool for filtering and formatting VCF files using Lua expressions, offering speed and flexibility comparable to existing tools.

STRling detects short tandem repeat expansions at known and novel loci using k-mer counting, achieving high sensitivity and base pair accuracy.

This paper introduces ConSpliceML, a machine learning model that improves the identification of deleterious splicing variants by integrating genetic constraint and alternative splicing predictions.

Echtvar compresses variant annotations to enable faster and more efficient SNP and indel analysis.

The paper establishes variant filtering guidelines for rare disease studies and estimates candidate variant yields for different inheritance modes using slivar software.

The paper introduces the D4 format and d4tools to efficiently store and analyze read depth data, achieving smaller file sizes and faster analysis speeds compared to existing formats.

Unfazed detects the parental gamete of origin for de novo mutations using read-based phasing and scales to large datasets.

Samplot creates visualizations for structural variant validation and includes a machine learning tool to reduce false positives across multiple sequencing technologies.

A C++ library plus 83 command line utilities and 44 scripts (vcflib) for filtering, transforming and computing population-genetic statistics on VCF files, with companion tools bio-vcf, slivar, cyvcf2 and hts-nim.

The paper presents a method to index and search raw alignments from thousands of samples to improve the classification of somatic and pathogenic structural variants.

GGD provides a fast, reproducible way to install standardized data recipes for genomics research, addressing challenges in data collection, transformation, and integration.

CaBagE is a method for enriching target DNA regions for long-read sequencing, achieving high coverage and enabling analysis of complex genomic loci in disease contexts.

The paper reports de novo structural mutation rates and gamete-of-origin biases using genome sequencing of 2,396 families, finding higher rates in autism cases and paternal origin dominance.

Somalier estimates sample relatedness and detects swaps in cancer and germline studies using efficient genome sketches without requiring germline variant calls.

The paper analyzes de novo mutations in three-generation CEPH families to show variability in mutation rates and identifies post-zygotic mosaicism as a source of de novo mutations.

The paper challenges the textbook view of germline mutations by showing that DNA damage and maternal age significantly contribute to mutation rates, with young fathers contributing more mutations than young mothers.

duphold improves structural variant call confidence by integrating depth-based annotations, enhancing specificity of deletion calls.

The paper creates a detailed map of constrained coding regions in the human genome using variation data from 123,136 individuals, identifying regions under strong purifying selection and enriched for pathogenic variants.

Bioconda provides a sustainable, comprehensive distribution of bioinformatics software using Conda for reproducible analysis across platforms.

SV-plaudit enables rapid manual curation of thousands of structural variants through a cloud-based framework with image visualization and collaborative reporting.

hts-nim provides a high-performance library for genomic analyses with a scripting-like syntax in Nim, addressing performance issues in scripting languages for large datasets.

Nanopore-only sequencing of GM12878 across 39 MinION flowcells (91.2 Gb) yielded an assembly with NG50 ∼3 Mb; adding 5× ultra-long reads raised NG50 to 6.4 Mbp.

GIGGLE is a search engine that identifies significant shared genomic loci between query features and large genomic datasets, improving data integration and hypothesis generation.

Mosdepth rapidly calculates genome-wide sequencing coverage from BAM or CRAM files, offering faster performance and flexible coverage profile options compared to existing tools.

Indexcov estimates whole-genome sequencing coverage to identify aberrant profiles, chromosomal anomalies, batch effects, and sample sex.