Geir Kjetil Sandve

This study creates a single-cell immune repertoire atlas in IBD patients, revealing coordinated immune states and identifying potential therapeutic targets.

This study benchmarks generative models for adaptive immune receptor repertoires, identifying their strengths and limitations in capturing biological features.

PEPE efficiently extracts multimodal protein embeddings from language models, enabling scalable and memory-efficient generation of large embedding datasets for downstream biological tasks.

This paper introduces immuneML, a framework for unsupervised machine learning in adaptive immune receptor repertoires, demonstrating its use in generative modeling, clustering, and dataset analysis.

The paper identifies TCR subsequence motifs associated with type 1 diabetes using deep learning and shows their enrichment in T1D individuals based on HLA risk alleles.

This study evaluates active learning strategies for antibody-antigen binding prediction in library-on-library settings, finding three algorithms that significantly improve efficiency and reduce experimental costs.

The paper quantitatively maps antigen specificity in immune repertoire embedding spaces, showing distinct regions for different antigen-specific sequences.

This study benchmarks mass spectrometry-based antibody sequencing methods, revealing biases related to sequence, concentration, protease, and bioinformatics tools.

This study shows that the composition of training data, especially negative examples, affects machine learning model performance and biological rule discovery in antibody-antigen binding.

This paper introduces LIgO, a software suite that simulates adaptive immune receptor repertoires with immune information to aid in developing and benchmarking AIRR machine learning methods.

The paper quantifies the plasticity of antibody developability by analyzing sequence and structure-based parameters in natural and human-engineered antibodies.

KAGE2 improves structural variant genotyping by using pangenomes, offering higher accuracy and speed compared to existing methods.

The paper introduces SNAIL, a normalization method that reduces false-positive co-expression associations in RNA-sequencing data, improving downstream network analysis.

GKAGE is a GPU-accelerated genotyper that improves the speed of SNP and short indel genotyping by an order of magnitude compared to CPU-based methods like KAGE.

BioNumPy enables efficient array programming on biological data using Python, outperforming vanilla Python and other packages while matching C/C++ speeds for common bioinformatics tasks.

The paper introduces immuneREF, a method to quantify immune repertoire similarity across health and disease, finding that healthy and diseased repertoires are more similar than previously thought.

CompAIRR rapidly compares adaptive immune receptor repertoires using exact or approximate sequence matching, significantly improving speed and memory efficiency compared to existing methods.

The paper evaluates ML models for adaptive immune receptor repertoire classification, finding that simple methods like L1-penalized logistic regression perform well even with rare immune signals.

KAGE is a fast and accurate alignment-free genotyper for SNPs and short indels using a pan-genome representation and Bayesian modeling.

immuneML provides an open-source ecosystem for reproducible and interoperable machine learning analysis of adaptive immune receptor repertoires, demonstrating its utility through benchmarking and novel methods.

Transformer attention is recast as the update rule of modern Hopfield networks, and the resulting DeepRC classifier handles immune repertoires with enormous instance counts and few informative sequences.

The paper evaluates two graph-based read mappers against a hybrid baseline approach, showing the baseline improves mapping accuracy to graph-based references.

immuneSIM simulates B- and T-cell receptor repertoires with tunable features to benchmark immunoinformatics tools.

NucBreak detects structural errors in genome assemblies using paired-end Illumina reads, showing high sensitivity and precision compared to existing tools.

The paper shows that the choice of metric significantly affects genomic co-occurrence analysis and proposes alternative approaches to guide this choice.

The paper introduces a method for calling ChIP-Seq peaks on graph-based reference genomes and demonstrates its ability to detect more motif-enriched peaks than MACS2.

The paper shows that ignoring assembly gaps in statistical tests for genomic feature co-localization can lead to false positives and inflated significance.

NucMerge improves genome assembly accuracy by integrating alternative assemblies and paired-end Illumina reads, outperforming existing tools in error correction and rearrangement detection.

The paper develops a method to predict direct TF-DNA interactions from ChIP-seq data and provides a resource of these interactions and cis-regulatory modules.

The paper highlights the importance of including genome build information in genomic track files and provides a tool to predict missing build versions.