Case-control analysis of single-cell RNA-seq studies

Petukhov Viktor

Khodosevich Lab, BRIC

viktor.petukhov@pm.me

Computational methods for single-cell analysis of brain disorders

Co-supervisor

  • Computational methods for single-cell biology
  • Harvard Medical School

Peter Kharchenko

Background

Main supervisor

  • Biology of neurodevelopmental disorders
  • scRNA-seq
  • University of Copenhagen

Konstantin Khodosevich

Structure

Overview of projects

Biological introduction

Projects: Conos

Projects: Epilepsy

Projects: Cacoa

Future directions

Slide complexity

Overview of projects

in prep.

preprint

package

published

authorship

Biology

Schizophrenia

Neuron Maturation

Macaque Vis Region

brain

Conos

Epilepsy

Cacoa

2019

2020

co-

co-

case-control

Baysor

SpaceTx

2022

spatial

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Our data: single-cell RNA-sequencing (2015)

DNA:

RNA molecules

Genes

}

Expression

levels

(3;

1;

5)

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Genes

DNA:

(3;

1;

5)

}

Expression

levels

t-SNE 1

t-SNE 2

*Kharchenko, P.V. The triumphs and limitations of computational methods for scRNA-seq. Nat Methods 18, 723–732 (2021)

Our data: single-cell RNA-sequencing (2015)

~10k cells x 20k genes

Genes

DNA:

(3;

1;

5)

}

Expression

levels

t-SNE 1

t-SNE 2

*Kharchenko, P.V. The triumphs and limitations of computational methods for scRNA-seq. Nat Methods 18, 723–732 (2021)

Our data: single-cell RNA-sequencing (2015)

Our data: multiple samples (2018)

Problem: multiple conditions (2020)

Control patients

Case patients

Problem: multiple conditions (2020)

Control patients

Case patients

Cells

Cell

types

Problem: multiple conditions (2020)

Control patients

Case patients

Cell

types

Cells

Cells

Genes

Cells

Genes

Cells

Genes

Cells

Genes

Cells

Genes

Cells

Genes

Cells

Genes

Cells

Genes

How to compare?

Conos

Epilepsy

Cacoa

Conos

Epilepsy

Cacoa

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Problem: batch-effect on multiple samples

*Integrating single-cell transcriptomic data across different conditions, technologies, and species, Butler A, et al. Nature Biotechnology, 2018

Problem: batch-effect on multiple samples

  • Individuals
  • Tissues
  • Species

Potential sources of differences:

  • Treatment
  • Environment
  • Disease
  • Protocol
  • Lab
  • ...

*Integrating single-cell transcriptomic data across different conditions, technologies, and species, Butler A, et al. Nature Biotechnology, 2018

Conos: clustering of networks of samples

Goal: develop a framework to analyze heterogeneous collections of samples without suffering from batch-effect

State of the art (Seurat, mnn): create a joint 'batch-corrected' expression space

Butler et al. Nat. Biotech’18

Haghverdi et al. Nat. Biotech’18

Conos: clustering of networks of samples

Goal: develop a framework to analyze heterogeneous collections of samples without suffering from batch-effect

State of the art (Seurat, mnn): create a joint 'batch-corrected' expression space

Problem: joint expression space removes important variation and distorts distributions

Conos: clustering of networks of samples

Problem: joint expression space removes important variation and distorts distributions

Solution: work with a joint topology (graph), not expression space

Conos: clustering of networks of samples

Pairwise alignment

Conos: clustering of networks of samples

Joint graph

Conos: clustering of networks of samples

Graph analysis

  • Clustering
  • 2D embedding
  • Propagating information between samples
  • Other graph processing

Conos: results

Annotation

Samples

Clusters

Conos: benchmarking

Projects

Conos

Epilepsy

Cacoa

Projects

Conos

Epilepsy

Cacoa

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Our data: multiple samples (2018)

Problem: multiple conditions (2020)

Healthy patients

Epilepsy patients

scRNA-seq to study Temporal Lobe Epilepsy

  • Temporal Cortex
  • 9 control vs 10 epilepsy samples
  • 117k nuclei

scRNA-seq to study Temporal Lobe Epilepsy

Goal: investigate molecular mechanisms of Temporal Lobe Epilepsy using single-cell data

State of the art: no scRNA-seq studies were done on Epilepsy data

Problem: how to measure changes between conditions?

Step 1: align and annotate

Step 2: identification of changed types

Gene expression analysis

Compositional analysis

%of nuclei

Step 2.1: compositional analysis

%of nuclei

Step 2.2: expression similarity

Step 2.2: expression similarity

Step 2.2: expression similarity

Step 2.2: expression similarity

Z = \frac{\overline{d}_{control} - d_{between}}{\overline{d}_{control}}

Step 2.2: expression similarity

Step 2.3: cell type ranking

Step 3.1: global structure of changes

developmental processes, neural circuit re-organization and neurotransmission

ion transport and glutamate signaling

protein transport to axons/dendrites

cell adhesion, ion transport and synaptic plasticity

regulation of neuronal morphogenesis

Step 3.2: local structure of changes

Step 3.2: local structure of changes

Control

Epilepsy

Step 3.2: local structure of changes

Step 3.2: local structure of changes

Projects

Conos

Epilepsy

Cacoa

Projects

Conos

Epilepsy

Cacoa

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Goal: develop a comprehensive set of methods for analysis of scRNA-seq case-control experiments

State of the art: no methods were published when we started, several competitors exist now

Problem: how to measure changes between conditions?

Case-control analysis of single-cell RNA-seq studies

Case-control analysis of single-cell RNA-seq studies

Compositional analysis

Gene expression analysis

Case-control analysis of single-cell RNA-seq studies

Case-control analysis of single-cell RNA-seq studies

Gene expression analysis

Compositional analysis

Cluster-based

Cluster-free

Control

Multiple sclerosis

Compositional analysis: cluster-based

Compositional analysis: cluster-based

*: CoDA significance

*: proportion significance

Compositional analysis: cluster-based

Compositional analysis: cluster-based

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Control

Multiple sclerosis

Compositional analysis: cluster-based

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Effect size

Significance

Expression analysis: cluster-based

EN L2-L3

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Expression analysis: sample structure

EN L2-L3

EN L2-L3

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Expression analysis: sample structure

Color by batch

Aggregated across all cell types

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Expression analysis: sample structure

Expression analysis: expression distance

separation

EN L2-L3

*Schirmer, L., Velmeshev, D., Holmqvist, S. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019)

Expression analysis: cluster-free

Expression analysis: cluster-free

Expression analysis: cluster-free

control

epilepsy

Expression analysis: cluster-free

control

epilepsy

Compare

Expression analysis: cluster-free shifts

Expression analysis: cluster-free DE

Expression analysis: cluster-free DE

Gene programs:

Expression analysis: cluster-free DE

Science reproducibility

>70% of publications had

major flows, compromising main results

Science reproducibility

>70% of publications had

major flows, compromising main results

Science reproducibility

  • Publish your code
  • Double-check your results
  • Don't get too excited

Future directions

  • Improving precision and power of methods
  • Including additional modalities
  • Including multiple conditions
  • Better handling covariates

Future work

Cacoa

Epilepsy

Conos

Overview

Introduction

Acknowledgements

Khodosevich Lab

Jonathan Mitchel

Ruslan Soldatov

Shenglin Mei

Evan Biederstedt

Navneet Vasistha

Konstantin Khodosevich

Rasmus Rydbirk

Irina Korshunova

Diego González

Katarina Dragicevic

Mykhailo Batiuk

Anna Igolkina

Peter Kharchenko

Acknowledgements

Thank you for your attention!

Petukhov Viktor

Khodosevich Lab, BRIC

Case-control analysis of single-cell RNA-seq studies

Made with Slides.com