department of english course descriptions fall 2012 - USC Dana and
The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of...
Transcript of The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of...
![Page 1: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/1.jpg)
![Page 2: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/2.jpg)
The power of single cells: Building a tumor immune atlas
Dana Pe’erDepartment of Biological Science
Department of Systems Biology
Columbia University
![Page 3: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/3.jpg)
The Precision Medicine Initiative
“Doctors have always recognized that every patient is unique. You can match a blood transfusion to a blood type — that was an important discovery. What if matching a cancer cure to our genetic code was just as easy, just as standard? What if figuring out the right dose of medicine was as simple as taking our temperature?” President Obama, 2015
Most “personalized medicine” efforts are focused on DNA,
but DNA will not “cut it”
![Page 4: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/4.jpg)
Controls Cases
TT
TT C
CCCC
C
Precision Medicine?
90% of the loci that associate with human traits and diseases are outside genes
Recent evidence supports that these fall in regions that regulate gene expression
DNA PHENOTYPERNA NETWORK
![Page 5: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/5.jpg)
Cells: Key intermediates from genotype to phenotype
When a genetic association is found:Which tissue dysfunctions?
![Page 6: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/6.jpg)
One Genome – Many Cell Types
ACCAGTTACGACGGTCA
GGGTACTGATACCCCAA
ACCGTTGACCGCATTTA
CAGACGGGGTTTGGGTT
TTGCCCCACACAGGTAC
GTTAGCTACTGGTTTAG
CAATTTACCGTTACAAC
GTTTACAGGGTTACGGT
TGGGATTTGAAAAAAAG
TTTGAGTTGGTTTTTTC
ACGGTAGAACGTACCGT
TACCAGTA
Expressed genes differ between cell types
The regulatory region of a gene differs between cell types!
Tissues contain many different cell types
3 billion letters
Gene expression is central to
understanding genetics and disease
![Page 7: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/7.jpg)
A Human Cell Atlas
Our genes are well mapped, but most of cell types remain unknown
Cells are basic biological units
Diseases are caused by malfunction of specific cell types.
Goal: Construct a comprehensive map of all cell types in our body
A cell atlas will be as empowering as the human genome map.
![Page 8: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/8.jpg)
8
A Geometric Approach to Phenotype
Emerging high dimensional single cell technologies: CyTOF, single-cell RNA-seq and MIBI allow us to characterize “phenotypic space”
Cell Phenotype: A configuration of multidimensional expression Defines a region in
“phenotypic space”
Data will consist of millionsof multi-parameter cells
Protein X
Pro
tein
Y
![Page 9: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/9.jpg)
viSNE map of healthy bone-marrow
Visualizing information derived
from many dimensions
t-SNE based Non-Linear Dimensionality Reduction Amir et. al. Nature Biotech 2013
![Page 10: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/10.jpg)
Cell phenotypes accumulate in
complex non-convex manifolds
![Page 11: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/11.jpg)
CD34 Expression
low high
Sample H1 Convert data to graph using
Jaccard metric
Graph approximates phenotypic manifold
Perform density estimation in the graph
Identify regions of phenotypic stability
Produce explicit labeling of subpopulations
A “social network” for cells
Levine*,Simonds* et. al. Cell 2015
![Page 12: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/12.jpg)
Phenograph, community detection
Community detection identifies densely interconnected node sets
Wi j: affinity function [ij coupling]
si: total affinity of i
ci: community assignment for i
2m: vol(W) [normalization]
Combinatorial optimization
Louvain method provides efficient heuristic (Blondel et al. J. Stat. Mech. 2008)
![Page 13: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/13.jpg)
PhenoGraph outperforms leading methods for subpopulation detection
Can run on 1 million cells in the same time it takes competing methods to
run on 80,000 cells
![Page 14: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/14.jpg)
Immunotherapy in Cancer
The miracle: 40% of metastatic melanoma patients showing “durable response” of many years
Success stories in many additional “bad cancers” including Lung, AML, Bladder, Glio-blastoma
Immunotherapy works for a small % of cancer patients, but when it works, it works
![Page 15: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/15.jpg)
Precision Cancer Medicine
Current efforts are based on “targeted therapy”, but
Cancer is so “smart and evolving”, simple drugs will not cut it.
Preexisting resistant clones present before treatment
Need smart and adaptive drug like our own immune system
Need: “big data” approaches to understand how immunotherapy can be extended to all patients
![Page 16: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/16.jpg)
Tumor Immune System Atlas
Goal: Characterize sub-populations in tumor immune ecosystem.
Challenge: Substantial unknown diversity.
A better understanding of tumor immune eco-system will aid the development strategies to activate it against the tumor.
Need thousands of CD45+ cells per tumor
![Page 17: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/17.jpg)
Cells
Hydrogel
beads
RT/lysis
Encapsulation
Klein*, Mazutis* et. al. Cell 2015
In-Drop Parallel Processing of RNA-Seq Librariesfrom >10,000 Individual Cells
Microfludic device can do 30,000 cells in one experiment
Tiny wells cut cost of reagents by 1000-fold
Highly scalable and inexpensive single-cell RNA-seq
![Page 18: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/18.jpg)
In-drop characterization of tumor immune cells in breast cancer
Data-Driven approach:
> 3000 CD45+ collected per tumor
Mean molecules per cell > 3500
Carr, Mazutis, Plitas with Rudensky lab, MSKCC
![Page 19: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/19.jpg)
tSNE 2D projection
CD45+ TILs from 4 breast cancers
Entire regions on the map are tumor specific
Are these differences real biology or technical effects?tumors
![Page 20: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/20.jpg)
Count Matrix2D projection of cells
(tSNE)Cells
Gen
es
Cluster 1 Cluster 2 Cluster 3
Single cell RNA-seq as imagined
![Page 21: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/21.jpg)
Problem: Single-cell RNA-seq data involves significant dropouts and library size variation
Observed Count MatrixCells
Gen
es
Cluster 1 Cluster 3
2D projection of cells (tSNE)
Need to Impute dropouts &
Normalize data
Typically sample only 5% of a cell’s
transcriptomeCluster 2
![Page 22: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/22.jpg)
Common Approach:Normalizing independent of cell types
To mean/median library sizeBASiC/ERCCs
Clustering Cells
Downstream Analysis
Problems: Dropouts not resolvedZeros remain zero! Removes biological
stochasticity specific to cell type
Leads to improper clustering
Biased results in downstream analysis
Normalization
Observed Count MatrixCells
Gen
es
![Page 23: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/23.jpg)
2D projection of cells (TSNE)
How can we impute expression in Single Cell RNA-seq data?
A
Prabhakaran*, Azizi* et.al, ICML 2016
![Page 24: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/24.jpg)
Idea 1: Impute dropouts based on expression in cells with same type
But we observe similarcells mostly express
Gene A
No expression of Gene A in a cell
A
![Page 25: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/25.jpg)
Idea 1: Impute dropouts based on expression in cells with same type
But we observe similarcells mostly express
Gene A
No expression of Gene A in a cell
Impute dropout in Gene A based on similar cells
A
![Page 26: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/26.jpg)
No significant inference based on
similar cells
Idea 2: Impute dropouts based on co-expression patterns
![Page 27: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/27.jpg)
Idea 2: Impute dropouts based on co-expression patterns
However Gene A always co-expressed with Gene B in
cells of same type
No significant inference based on
similar cells
27
![Page 28: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/28.jpg)
Idea 2: Impute dropouts based on co-expression patterns
Impute dropout in Gene Abased on Gene B
No significant inference based on
similar cells
However Gene A always co-expressed with Gene B in
cells of same type
![Page 29: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/29.jpg)
Imputing & Normalization
Histogram of library size in example datasetFrom Zeisel, Science 2014
In addition to imputing dropouts, we need to normalize data by library size
![Page 30: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/30.jpg)
Problem with Global Normalization
Example Housekeeping Gene
Cells with different sizes have very different total number of transcripts
High chance of Dropouts in smaller cells
![Page 31: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/31.jpg)
Problem with Global Normalization
Dropout not resolved
Spurious Differential Expression
![Page 32: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/32.jpg)
Idea: Different normalization for each cell type
Problem: We don’t know cell types
Need to infer cell clusters
![Page 33: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/33.jpg)
Approach: Simultaneous inference of clusters and imputing parameters
iterative learning
Imputing &Normalization
Clustering Cells
x 3
x 1
x 0.75
x 1
![Page 34: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/34.jpg)
Cell-specific Parameters
for Imputing & Normalization
Assignment of cells to clusters
Parameters characterizing each cluster iterative
learning
Approach: Simultaneous inference of clusters and imputing parameters
x 3
x 1
x
0.75
x 1
![Page 35: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/35.jpg)
Modeling: Clusters of Cellsusing a Bayesian Mixture Model
Cells
Ge
ne
s
Cluster 1 Cluster 2 Cluster 3
Ideal Count Matrix (normalized)
![Page 36: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/36.jpg)
Model distributions of Log of countsfor each gene per cell type as a
Gaussian distribution
Cluster 1 Cluster 2 Cluster 3
One
gene
Modeling: Clusters of Cellsusing a Bayesian Mixture Model
Cells
Ge
ne
s
Cluster 1 Cluster 2 Cluster 3
Ideal Count Matrix (normalized)
![Page 37: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/37.jpg)
Cluster 1 Cluster 2 Cluster 3
Each gene: Mixture of Log-Normal Models
Modeling: Clusters of Cellsusing a Bayesian Mixture Model
Cells
Ge
ne
s
Cluster 1 Cluster 2 Cluster 3
Ideal Count Matrix (normalized)
One
gene
![Page 38: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/38.jpg)
Cluster 1
Cluster 2Cluster 3
Modeling all genes together: Mixture of Multivariate Log-Normals
To also take advantage of co-expression patterns in learning clusters
Two genes
Modeling: Clusters of Cellsusing a Bayesian Mixture Model
Cells
Ge
ne
s
Cluster 1 Cluster 2 Cluster 3
Ideal Count Matrix (normalized)
![Page 39: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/39.jpg)
Generative Model with Technical Variation W
itho
ut
Te
chnic
al
Variatio
n
With T
echnic
al
Variatio
nLatent counts which we
want to recover
Observation
![Page 40: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/40.jpg)
Cluster-specific parameters Cell-specific
parameters
Observed gene
expression per cell j
BISCUIT (Bayesian Inference for Single Cell ClUstering and ImpuTing)
![Page 41: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/41.jpg)
BISCUIT (Bayesian Inference for Single Cell ClUstering and ImpuTing)
Cluster-specific parameters Cell-specific
parameters
Priors set based on observed lib size distribution
Hyper-parameters
Hyper-priors
Global Distributions of Observed Data
Observed gene
expression per cell j
![Page 42: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/42.jpg)
Inference Algorithm
Sampling technical variation parameters
Sampling assignment of cells to clusters
using Chinese Restaurant Process
(CRP)
Sampling cluster-specific parameters
Sample hyper-parameters
Gibbsiterations scaling mean, cov per cell
Also allows estimating the
number of clusters
Estimate hyper-priors based on Data
Parallel Sampling from derived conditional posterior distributions: P(parameter| data, other parameters)
![Page 43: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/43.jpg)
BISCUIT clusters Cells
Gen
es Observed Data
Cluster-dependent Imputing & Normalizing
![Page 44: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/44.jpg)
Sort cells into clusters
BISCUIT clusters Cells
Gen
es Observed Data
Cluster-dependent Imputing & Normalizing
![Page 45: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/45.jpg)
BISCUIT clusters Cells
Gen
es Observed Data
Cluster-dependent Imputing & Normalizing
![Page 46: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/46.jpg)
Cells
Gen
es Observed Data
BISCUIT clusters & parameters
Cluster-dependent Imputing & Normalizing
![Page 47: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/47.jpg)
Cells
Gen
es Observed Data
Cluster-dependent Imputing & NormalizingBISCUIT clusters & parameters
Impute & Normalize
With a linear transformation
![Page 48: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/48.jpg)
Cells
Gen
es Observed Data
Cells
Gen
es
Cluster 1 Cluster 2 Cluster 3
Imputed Data
Cluster-dependent Imputing & NormalizingBISCUIT clusters & parameters
Impute & Normalize
With a linear transformation
![Page 49: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/49.jpg)
Performance: Testing on neuron single cell data 3005 mouse cortex cells from Zeisel et al., Science 2015 Deep coverage (2 million reads per cell) gives good ground truth for 7 Cell types. No prior information used: selected 558 genes with largest standard deviation across cells
F-score: 0.91
![Page 50: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/50.jpg)
Comparing: Biscuit to other methodsF-score: 0.91 F-score: 0.79
F-score: 0.74 F-score: 0.61F-score: 0.5 for 67% of cells
![Page 51: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/51.jpg)
Reminder: Breast TIL data before Biscuit
Skews data, non-overlapping cells across tumors
Unclear structure of cell types, mostly distinguishes myeloid from lymphoid cells
tumors
![Page 52: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/52.jpg)
Breast cancer TIL data after Biscuit
Most of the tumor specific regions vanish
Most of the map includes cells from all 4 tumors
12,000 Cells, ~3000 molecules per cell
4 tumors
![Page 53: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/53.jpg)
Breast cancer TIL data after Biscuit
Tregs (FOXP3)
T-cells CD4+ (CD3, CD4)
T-cells CD8+(CD3D, CD8A, CD8B)
NKs (NKG7)
Bcells (CD79A, CD19)
Myeloid (CD14+,CD81+, APOE-)
Monocytes(CD14,CD68)
DCs (CD1C)
mast cells
Myeloid (CD14+,CD81+, APOE+)
Exhausted T-cells
![Page 54: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/54.jpg)
Patient 1 Stronger
covariance between ICOS and CTLA4
Weaker
covariance between OX40 and GITR
Patient 2
Patient 3 Patient 4
Patient specific differences in co-variation structure
These are the co-receptors that are targeted by immunotherapy
![Page 55: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/55.jpg)
Summary for Biscuit
We introduce BISCUIT: iteratively clusters and normalizes single-cell RNA-seq data based on different cell types.
hierarchical Bayesian mixture model with an efficient Gibbs sampler for inferring cell-specific parameters.
imputes dropout gene expression values.
We constructed a cell atlas of the tumor immune system: Captured a rich diversity of tumor immune cell types
Cancer specific differences in co-receptor patterns that can guide combinatorial immunotherapy
(releasing multiple breaks).
![Page 56: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/56.jpg)
Once we have the parts we can ask how these interact Effect of tissue context, organ site, cancer/healthy? What happens in response to drug?
![Page 57: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/57.jpg)
Longitudinal Studies: How does the tumor ecosystem change under drug perturbation?
both cell state and cell to cell interactions How does this differ between responders and non-responders? Our measurements are genome-wide, mechanism!!
![Page 58: The power of single cells: Building a tumor immune atlas€¦ · Dana Pe’er Department of Biological Science Department of Systems Biology Columbia University. The Precision Medicine](https://reader033.fdocuments.in/reader033/viewer/2022052014/602b29183033a5294879c94a/html5/thumbnails/58.jpg)
Acknowledgements
Ambrose CarrLinas Mazutis
Rami Eitan
Josh Nainys
Sasha Rudensky (MSKCC)
George Plitas
Casia Konopac
Patients
Elham AziziSandhya Prabhakaran
Kristy Choi
Manu Setty
Vaidas Kiseliovas
Jacob Levine
Zakary Posewitz