Mika Gustafsson

Protein-protein interaction priors shape biologically coherent latent spaces for causally concordant cross-omic translation

This paper introduces a PPI prior to guide autoencoder embeddings, improving biological coherence and downstream task performance in multi-omic data.

System-level health profiling from blood DNA methylation with explainable deep learning

The paper develops an explainable deep learning framework to predict health scores from blood DNA methylation, showing strong predictive accuracy and alignment with biological processes.

Representational Learning from Healthy Multi-Tissue Human RNA-seq Data such that Latent Space Arithmetics Extracts Disease Modules

This study uses a variational autoencoder trained on healthy human RNA-seq data to predict disease gene expression changes and identify drug targets through latent space arithmetic.

A comprehensive mechanistic model of adipocyte signaling with layers of confidence

The paper develops a method to build a comprehensive adipocyte signaling model by iteratively adding phosphosites with confidence levels, achieving high predictive accuracy for inhibitor data.

Data-driven representations using deep network-coherent methylation autoencoders to identify robust disease and risk factor signatures

This study uses deep autoencoders to analyze DNA methylation data and identifies disease and risk factor signatures related to aging, smoking, and disease.

White-box Deep Neural Network Prediction of Genome-Wide Transcriptome Signatures

The paper predicts human transcriptomic profiles from transcription factor expression using a deep neural network, finding that 1,600 TFs explain >95% of variance in 25,000 genes.

ComHub: Community predictions of hubs in gene regulatory networks

ComHub predicts hub transcription factors in gene regulatory networks by averaging predictions from multiple network inference methods, showing robust performance on benchmark datasets.

A validated generally applicable approach using the systematic assessment of disease modules by GWAS reveals a multi-omic module strongly associated with risk factors in multiple sclerosis

The paper benchmarks disease module identification methods using GWAS and validates a multi-omic module strongly associated with multiple sclerosis risk factors.

A validated strategy to infer protein biomarkers from RNA-Seq by combining multiple mRNA splice variants and time-delay

The paper develops a mathematical model combining mRNA splice variants and time-delay to predict protein abundance from RNA-Seq, improving biomarker discovery in immune diseases.

Deriving Disease Modules from the Compressed Transcriptional Space Embedded in a Deep Auto-encoder

The paper uses a deep auto-encoder to discover disease-related gene modules from transcriptional data without relying on prior biological networks, finding significant enrichment of GWAS genes in the third layer.

LiPLike: Towards gene regulatory network predictions of high-certainty

LiPLike improves gene regulatory network predictions by reducing false positives and increasing accuracy compared to existing methods.

ModuleDiscoverer: Identification of regulatory modules in protein-protein interaction networks.

ModuleDiscoverer identifies regulatory modules in protein-protein interaction networks using gene expression data and validates them with SNP data from GWAS.

LASSIM - a network inference toolbox for genome-wide mechanistic modeling

LASSIM infers genome-wide gene regulatory networks using mechanistic ODE models, demonstrating its effectiveness in modeling Th2 differentiation with integrated data and siRNA experiments.