Single cell Omics

From Colettapedia
Revision as of 15:50, 23 May 2019 by Colettace (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Mike Kelly talk


  • Part of the standard repertoire of biological research techniques
  • Look at heterogeneity
  • avoids caveat of bulk averaging
  • heterogenous tissue
  • cell number
  • developmental stages
  • allows inference of dynamic processes
  • multiple asynchronous states of a cell states
  • transcriptional profile that happens across those time stages
  • interrogate potential mechanisms at cellular resolution

Sample Prep

  1. Solid tissue
  2. dissociation
  3. single cell isolation
  4. generate cdna library
  5. amplify cdna productcuts ususlally PCR
  • droplet-based gel cell barcode
  • plate-based sc-Seq using flow cytometry - allows gating to target cells of interest
  • What is the RNA quality? We don't know. Fixed cells? time since extraction can have a big affect on data quality
    • full length transcript allows get isoform analysis
  • 10xgenomics prep guidelines
    • single cell atac-seq important to. start with their protocols and buffers
  • Try for 90% viability threshold
  • minimize dead cells nucleic acids inhibitors of reverse transcription
  • Miltenyi dead cell removal
  • single NUCLEI sequencing - option for difficult to dissociate tissue, strongly-adherent, fragile cells like neurons, or solid frozen samples
    • More agnostic to variability of selection bias
    • But lower RNA content
    • higher introns retained

Add-on Modalities

  • VDJ sequencing solution from 10x genomics
    • TCR-VCR sequencing
    • TCR clonotypes grouped by t-SNE
  • Antibody feature barcoding
    • cell surface protein coding
    • similar to FACS
    • limited to surface proteins, can't get at transcription factors with this technique
  • Sample multiplexing
    • same as tagging sample subtypes, but have different barcoding
    • can associate which original sample they came from
    • run all the samples together in the sequencing run, reduces technical variation for biological replicates, more statistical power

Analysis Toolkits

Public datasets

Nan Ping talk

  • innate immunity - non specific
  • adaptive immunity - b cells antibody, CD4 hemper cells and CD8 killer cells
  • functional decline in age
  • naive t cells - then expansion and proliferation into effector cells, becomes memory cells. 7 different types of t-cells known
  • What type of lymphocyes are reduced by age - naive
  • how many CD8 sub populations are there?
  • Age increase in number of cells, or expression level within a cell, or both?

Jesse Gillis talk

  • Composition across samples explains huge proportion of variation in phenotype
  • controlling for batch effects
  • brain cell census network (BICCN)
  • why does this work at all. non-trivial that this data can be assessed, that methods can work.
  • What creates , is that there is independence, replicability
  • tabula muris consortium 2018
  • shallow-depth works: low n sample, can still group them in biologically meaningful way
  • low-dimensional 20,000 gene, many genes are proxies for each other, coexpression, t-sne captures this reasonably well (Crow & Gillis Trends in genomics 2018)
  • variation in expression variation is more biological than technical effects
  • Expression profiles are cvery comparable across datasets, can look for analog gene
  • cell types is the building block on which variation

Brain organization

  • cell types as basic components of brain circuitry
  • how do we recognize a new cell type? electrophysiological properties
  • rely on redundancy of coexpression
  • median number of genes detected per cell


  • t-SNE distances don't matter because the space is warped, global distances aren't meaningful at all
  • Is batch effect confounding biological and technical variation
  • MetaNeighbor sees through batch effects (Crow 2018)
  • replicability of 20 subclasses and 61 clusters of cell types replicable across datasets
  • predict what cell I am, stratification at the dataset level, learn something robust to batch.
  • pick one population, train, stratification on the class of variability, generates robust results
  • Need roughly 50,000 cells to get resolution to see replicable clusters of cell types
    • don't need a million cells for cell type experiment, but perhaps need that for other types of experiments
  • imputation may rescue depth, interact with batch effects? There is no right analysis method, model that you use to answer question. Judge whether two expression cells, renders non-independent meaningless co-expression. Cell profile vs. transcriptional profile. Analyze cell independently, imputation makes it harder to understand how artifacts drive results.
  • Cells are dying, nuclei are robust to dying, are you just measuring how much cells are dying. all cell functions are not within the nuclei. Nuclei are missing real biological variation
  • methylation is correlated with expression
  • atac-seq
  • brain tumor gives good expression from frozen tissue - tissue/sample/brep dependent. Is it a brain specific result? Nuclei vs. cell we'll have to think about biological

Sara Linker talk

  • old methods are prformed at heat over a long period of time
  • nuclear is realtively quick on ice
  • Newer 10x genomics new methods get better clustering with single nucleus sequencing
  • transcriptional component of memory
  • focus on dentate gyrus in hippocampus, contextual memory.
  • memory is not a bulk event, not need to reactivate every cell to get context, strong activating gene, then separated heterogeneity, some neurons are primed to be reactivated.
  • whole cell get nuclear and cytoplasmic, lose rnas that are dendritic, but heterogenous loss of dendrites
  • nuclei homogenous loss of dendrites
  • look at activity induced expression
  • FOS can go form 1% to 90% activation
  • but no difference
  • nuclear rna seq method could maintain the genes we are looking at.
  • experiment 1, actuvate neurons naturally rather than through seizure, sor out fod + and neg
  • see difference in expression in early activation genes, mitigates effects of dissociation procedures
  • baseline in the home cage, immediate early signiture, late signiture
  • FOS on at 1hr, off at 4 h, but ARC stays on
  • multiple waves of transcirption
  • temporal profile to discriminate between room a and room c
  • same room habituates, subset of neurons will fire again
  • different room, different neurons firing

Eddie Lee

  • "sentenel case"
  • Tau pathology tight correlation, plus beta amyloid
  • starts in medial temporal lobe
  1. A-beta accumulation
    • APOE risk allele
  2. microglial activation (inflammatory response? loss favor, but in resurgence)
  3. neurofibulary tangles
  4. dementia
  5. death
  • autosomal dominant mutations (<5% cases)
    • APOE
    • TREM2
  • non-mendelian form of AD >95% cases, sporadic, >65%
  • clearing beta amyloid show no benefit
  • genome-wide associated risk alleles
  • PENN brain bank
  • microglia transcriptomes change very rapidly, hours, maybe minutes
  • DESC: deep embedding: autoencoder, find the features to classify. Look for homogenous mix of samples in t-SNE clusters
  • Pseudotime Trajectory Analysis (PAGA)
  • microglial phenotypes : pathology x Genetics
  • when cells are dying microglia come in to clean up the mess
  • With aging, neurons experience transcriptional noise?