Ole Winther

DeepCDS: Ab initio coding sequence prediction in prokaryotic short nucleotide sequences

DeepCDS improves coding sequence prediction in short prokaryotic sequences using deep learning and protein language model embeddings, outperforming existing methods across various sequence lengths and error profiles.

DeepTMHMM2 enables accurate prediction of transmembrane protein topology and subcellular location

DeepTMHMM2 predicts transmembrane protein topology and subcellular location, including re-entrant regions and interfacial helices, and identifies localization across 17 membranes.

SpatialFormer: Universal Spatial Representation Learning from Subcellular Molecular to Multicellular Landscapes

SpatialFormer integrates spatial and gene expression data to improve tasks like batch correction and cell-type annotation in spatial transcriptomics.

Predicting the subcellular location of prokaryotic proteins with DeepLocPro

DeepLocPro is a subcellular location predictor for prokaryotic proteins, outperforming PSORTb 3.0 on benchmark metrics using experimentally verified data.

DeepLocRNA: An Interpretable Deep Learning Model for Predicting RNA Subcellular Localization with domain-specific transfer-learning

DeepLocRNA predicts RNA subcellular localization using transfer learning from pre-trained RBP-binding models, achieving state-of-the-art performance across human and mouse RNA.

GraphPart: Homology partitioning for biological sequence analysis

GraphPart partitions biological sequences to avoid closely related sequences in different partitions, retaining more sequences than reduction methods while maintaining homology separation quality.

Towards In-Silico CLIP-seq: Predicting Protein-RNA Interaction via Sequence-to-Signal Learning

RBPNet predicts CLIP crosslink count distributions from RNA sequences, outperforming existing methods and enabling variant impact scoring.

Deep integrative models for large-scale human genomics

The paper introduces a deep learning framework for polygenic risk score prediction, showing improved performance over traditional methods for various diseases.

MembraneFold: Visualising transmembrane protein structure and topology

MembraneFold integrates protein structure and topology prediction to visualize transmembrane proteins, offering a web server for fast and accurate analysis.

Identifying endogenous peptide receptors bycombining structure and transmembrane topologyprediction

The paper uses AlphaFold-Multimer and DeepTMHMM to identify cognate receptors for secreted endogenous peptides by prioritizing interactions with high confidence metrics and filtering biologically implausible ones.

Improved prediction of blood biomarkers using deep learning

This study uses deep learning to predict blood and urine biomarkers from genetic data, outperforming linear methods and capturing non-linear and epistatic effects.

Transfer learning identifies sequence determinants of regulatory element accessibility

This paper introduces ChromTransfer, a transfer learning method that improves cell-type specific chromatin accessibility prediction from DNA sequences using pre-trained models.

DeepTMHMM predicts alpha and beta transmembrane proteins using deep neural networks

Combining an ESM-1b encoder with a conditional random field decoder, DeepTMHMM classifies alpha-helical and beta-barrel transmembrane proteins and their topology, outperforming most compared tools, including on signal-peptide cleavage sites.

SignalP 6.0 achieves signal peptide prediction across all types using protein language models

SignalP 6.0 predicts all five types of signal peptides with improved accuracy and applies to metagenomic data.

A comparison of tools for copy-number variation detection in germline whole exome and whole genome sequencing data

The paper benchmarks CNV detection tools on WGS and WES data using clinical standards and reference samples, finding that combining multiple tools improves recall but precision remains a challenge.

Binning microbial genomes using deep learning

This paper introduces a deep learning method using variational autoencoders to improve microbial genome binning from metagenomic data, achieving higher precision and completeness in reconstructed bins.

scVAE: Variational auto-encoders for single-cell gene expression data

scVAE is a variational auto-encoder for scRNA-seq data that avoids preprocessing and improves clustering by using raw counts and latent space clustering.

Language modelling for biological sequences - curated datasets and baselines

The paper analyzes properties of protein datasets for language modelling and evaluates n-gram and RNN models, finding that organism domain and data quality affect performance while completeness has little impact.

Single mRNP analysis by super-resolution microscopy and fluorescence correlation spectroscopy reveals that small mRNP granules represent mRNA singletons

The paper uses super-resolution microscopy and fluorescence correlation spectroscopy to show that small mRNP granules contain single mRNAs and RNA-binding proteins, suggesting they are mRNA singletons.

Detecting Novel Sequence Signals in Targeting Peptides Using Deep Learning

The paper introduces TargetP 2.0, a deep learning method to identify N-terminal sorting signals in proteins, revealing insights into the importance of the second residue in targeting peptides.

NetSurfP-2.0: improved prediction of protein structural features by integrated deep learning

NetSurfP-2.0 improves prediction of protein structural features with higher accuracy and faster runtime than previous versions.