Dmitry Kobak

This paper introduces a compound model for single-cell RNA-seq data without UMIs, using Pearson residuals to improve gene selection and embeddings in Smart-seq2 datasets.

The paper develops a hybrid modeling approach to link ion channel gene expression to the electrophysiological properties of cortical neuron types using statistical and mechanistic models.

BiPCA improves PCA for omics count data by adaptively rescaling noise, enhancing biological interpretation and reducing batch effects across multiple modalities.

The paper challenges claims that t-SNE and UMAP embeddings of single-cell data are misleading, showing they can still provide biologically relevant information despite not preserving high-dimensional distances.

The paper explores a spectrum of visualization methods between t-SNE and UMAP for single-cell data, showing how varying a single parameter affects the representation of local vs. continuous structures.

The paper identifies distinct feedback circuit motifs in cortico-cortical pathways, showing different functional dynamics in motor-sensory and visual feedback circuits.

The paper generates a multimodal cell census and atlas of the mammalian primary motor cortex, revealing a unified molecular genetic landscape and cross-species taxonomy of cell types.

This paper compares normalization methods for single-cell RNA-seq data, showing that analytic Pearson residuals outperform other approaches in identifying variable genes and dimensionality reduction.

This paper introduces sparse reduced-rank regression for visualizing relationships between paired multivariate datasets, showing improved performance over existing methods in Patch-seq data.

The paper shows that UMAP's apparent superiority over t-SNE in preserving global structure is due to different initialization methods, not the algorithm itself.

The paper combines morpho-electric and transcriptomic data from mouse motor cortex neurons to show that transcriptomic types within families form continuous morpho-electric landscapes rather than discrete entities.

This study shows that motor skill learning reduces motor variability and increases planning horizon, with an optimal control model capturing these changes.

The paper evaluates different neuronal morphology representations for cell type discrimination using curated datasets of mouse neurons, finding 2D density maps and morphometric statistics to be the most effective.

The paper characterizes cell types and circuit organization within neocortical radial clones using single-cell RNA-sequencing and electrophysiological recordings, finding that clones contain diverse cell types and exhibit specific connectivity patterns.

The paper describes a protocol to improve t-SNE visualizations in single-cell transcriptomics by using PCA initialization, high learning rate, and multi-scale kernels.

The paper identifies distinct cell types and circuit organization in layer 4 of primary sensory areas in adult mouse neocortex, challenging the canonical microcircuit model.

The paper investigates how brain state affects the geometry of population activity in rat auditory cortex using sound stimuli and finds state-dependent changes in coding properties and noise structure.