Mile Šikić

A complete diploid human genome benchmark for personalized genomics

T2T-HG002v1.1, a complete diploid HG002 assembly lacking detectable errors over 99.4% of its length, enables genome-based benchmarking of assemblies, sequencing reads, and phased variant calls.

Binary node clustering via contrastive learning for haplotype phasing in de novo genome assembly

This paper introduces grapHiC, a contrastive learning framework for reference-free haplotype phasing in de novo genome assembly, achieving results comparable to state-of-the-art methods.

GnnDebugger: GNN based error correction in De Bruijn Graphs

GnnDebugger uses Graph Neural Networks to correct errors in De Bruijn Graphs, improving genome assembly accuracy for diploid genomes with low coverage.

A near-complete human pancreatic cancer genome

This paper presents near-complete, haplotype-resolved assemblies of a pancreatic cancer genome and normal tissue, identifying extensive somatic variants and epigenetic changes in previously excluded regions.

Metaxa: A Transformer-Based Deep Learning Model for Taxonomic Classification of Long Nanopore Reads

Metaxa is a transformer-based model for classifying long Nanopore reads, outperforming existing tools on out-of-sample data and across different sequencing chemistries.

Campolina: A Deep Neural Framework for Accurate Segmentation of Nanopore Signals

Campolina improves nanopore signal segmentation accuracy using deep learning, outperforming traditional methods on real datasets.

DipGNNome: Diploid de novo genome assembly with geometric deep learning and beam-search

DipGNNome introduces a deep learning framework for diploid genome assembly using graph neural networks and beam-search, achieving results comparable to state-of-the-art methods.

MADRe: Strain-Level Metagenomic Classification Through Assembly-Driven Database Reduction

MADRe improves strain-level metagenomic classification by combining assembly-driven database reduction with probabilistic read mapping, achieving higher precision and fewer false positives than existing tools.

HomRank: Homogeneous RNA Ranking for 3D Structure Evaluation

The paper introduces HomRank, a ranking method for RNA 3D structure assessment that improves accuracy over existing methods on benchmark datasets.

RNA-xLSTM: Evaluating xLSTM as an Alternative Foundation to Transformers in RNA Modeling

The paper evaluates xLSTM as an alternative to transformers for RNA sequence modeling, comparing it to RNA-FM and RiNALMo on secondary structure and splice site prediction.

A Complete Telomere-to-Telomere Diploid Reference Genome for Indian Population

The paper presents a high-quality, gapless diploid reference genome for an Indian population, revealing extensive structural variation and improved performance for genomic analysis in South Asian samples.

Geometric deep learning framework for de novo genome assembly

This paper introduces GNNome, a geometric deep learning framework for de novo genome assembly that improves contiguity compared to existing tools across multiple species.

A systematic benchmark of Nanopore long read RNA sequencing for transcript level analysis in human cell lines

This study benchmarks Nanopore long read RNA sequencing in human cell lines, showing improved isoform resolution compared to short reads.

The Hitchhiker's Guide to Sequencing Data Types and Volumes for Population-Scale Pangenome Construction

This study evaluates optimal long-read data types and volumes for chromosome-level haplotype-resolved pangenome assembly, providing guidelines for sequencing platform selection and coverage.

A Comparative Review of Deep Learning Methods for RNA Tertiary Structure Prediction

This study evaluates six deep learning models for RNA 3D structure prediction across multiple datasets, identifying their strengths and limitations.

Rockfish: A Transformer-based Model for Accurate 5-Methylcytosine Prediction from Nanopore Sequencing

Rockfish improves 5-methylcytosine detection from Nanopore sequencing with higher accuracy and efficiency compared to existing methods.

Telomere-to-Telomere Phased Genome Assembly Using HERRO-Corrected Simplex Nanopore Reads

Deep-learning correction of ultra-long nanopore Simplex reads lifts read accuracy up to 100-fold and yields phased telomere-to-telomere human assemblies without a second sequencing technology.

Comprehensive benchmarking of metagenomic classification tools for long-read sequencing data

The paper benchmarks metagenomic classification tools for long-read sequencing data, finding that general-purpose mappers outperform classification tools in accuracy but are slower, while kmer-based tools are faster but less accurate.

Raven: a de novo genome assembler for long reads

Raven is a de novo genome assembler for long reads that offers high accuracy, speed, and low memory usage across various genomes.

Yet another de novo genome assembler

This paper introduces a new de novo genome assembler for long uncorrected reads, emphasizing speed and memory efficiency for large genomes.

Graphmap2 - splice-aware RNA-seq mapper for long reads

Graphmap2 improves transcript detection and exon boundary recognition using long-read RNA-seq data, outperforming existing tools in accuracy and recall.

Nanopore sequencing enables high-resolution analysis of resistance determinants and mobile elements in the human gut microbiome

This paper introduces a hybrid metagenomic assembler that improves assembly quality and enables high-resolution analysis of resistance determinants and mobile elements in the human gut microbiome.

Evaluation of tools for long read RNA-seq splice-aware alignment

This study evaluates RNA-seq splice-aware alignment tools for long-read sequencing technologies, showing that some tools perform well with error correction.

RNA Transcriptome Mapping with GraphMap

The paper introduces a modified RNA mapper based on GraphMap for third-generation sequencing data, showing improved mapping quality over existing tools.

SWORD - a highly efficient protein database search

SWORD is a protein database search tool that outperforms BLASTP in speed and sensitivity, offering faster execution and optimal alignments for candidate sequences.