Alexander Schönhuth

ANIMA predicts protein-protein interactions across species using an AI model trained on diverse animal species, achieving high accuracy in under-researched species.

HapFold improves haplotype reconstruction for diploid genomes by combining graph-based and sequence-based scaffolding methods, achieving high accuracy and contiguity.

PanTax improves strain-level metagenomic classification by using pangenome graphs, outperforming existing methods in precision and recall.

This paper develops a deep learning model using SNP microarray data to predict ALS risk with high accuracy across diverse populations and genotyping platforms.

PanSpace is a convolutional neural network-based tool for fast and scalable indexing of bacterial genome databases, reducing disk space and improving query speed.

DiVerG is a scalable data structure and algorithm for efficiently calculating distances in sequence graphs, outperforming PairG in speed and space efficiency for genomic applications.

PhenoEncoder improves genomic data compression by integrating phenotype information, enhancing downstream classification performance in complex trait studies.

PanTax improves strain-level taxonomic classification of metagenomic data by using pangenome graphs, outperforming existing methods in precision and recall.

HyLight improves strain-aware metagenome assembly by combining TGS and NGS data, achieving higher accuracy and lower costs than existing methods.

The paper introduces a method to compress genetic data using autoencoders that leverage linkage disequilibrium structure, achieving high reconstruction accuracy while reducing data size.

HERO improves error correction in long reads by combining de Bruijn and overlap graphs, reducing indel and mismatch errors by 65% and 20% respectively.

This paper introduces DiseaseCapsule, a capsule network approach for predicting complex genetic diseases from genotypes, achieving high accuracy on ALS and Parkinson's disease.

VeChat corrects errors in long-read sequencing data by using variation graphs instead of sequence-based consensus, reducing errors in PacBio and ONT reads compared to existing methods.

StrainXpress improves strain-aware metagenome assembly by reconstructing strain-specific genomes from short reads, outperforming existing methods in coverage and accuracy.

Strainline assembles full-length viral haplotypes from noisy long reads without a reference genome, outperforming existing methods in simulated and real datasets.

ProSolo improves variant calling accuracy in single-cell DNA sequencing by addressing amplification bias and errors through a probabilistic model and joint modeling with bulk data.

This paper introduces phasebook, a method for haplotype-aware diploid genome assembly from long reads, showing improved haplotype coverage compared to existing methods.

MarkerCapsule improves single cell typing by using capsule networks, requiring less labeled data and providing biologically meaningful gene activity patterns, while outperforming existing methods.

OGRE clusters metagenomic reads into species-specific bins efficiently for large datasets without computational limitations.

This paper introduces a statistical model to improve sensitivity and control false discovery rate in somatic indel discovery by accurately quantifying uncertainties.

VG-Flow enables strain-aware full-length genome assembly from mixed samples using a de novo approach that improves contiguity and haplotype abundance estimation.

The paper proposes a feature selection method to identify key miRNAs for cancer classification, finds a signature of 100 miRNAs that accurately classify 29 cancer types, and discovers 15 new potential biomarkers through literature meta-analysis.

MALVA is a mapping-free genotyping method that efficiently handles multi-allelic SNPs, indels, and high-density variants with high accuracy and speed.

The paper introduces a method for full-length de novo viral quasispecies assembly using variation graphs, improving contiguity and accuracy over existing methods.

This paper develops a deep learning approach to predict ALS from genotype by leveraging genome structure, showing better performance than other methods.

POLYTE generates haplotype-resolved de novo assemblies for diploid and polyploid genomes using an overlap graph-based method, outperforming existing approaches in accuracy and efficiency.

SAVAGE reconstructs individual haplotypes of intrahost viral strains without a reference genome, outperforming existing methods on simulated and real data.

This paper introduces a latent variable model to improve sensitivity and control false discovery rate in somatic indel discovery, showing significant improvements in recall for deletions.

Re-aligning reads to both alleles gives WhatsHap the lowest PacBio switch error rates (full coverage: 0.24% real, 0.02% simulated) and better indel phasing than hapCUT and phASER.