Syzygy Forces

watch this presentation live on your phone

Christakis & Fowler NEJM 2007

Zappasodi et. al. Cancer Cell 2018

quantitative

qualitative

eff

T

B

PD-1

CD4+

Foxp3-

T

fh

-like

CTLA

T

reg

(...maybe)

before/after

Zappasodi et. al. Cancer Cell 2018

from this:

to this:

current models = quantitative + qualitative layers

data sharing: why would I? ...no shared model

network topology helps us understand connectivity of cells

Why network analysis?

healthy immune network

what is a baseline "healthy" immune system?

is TBNK the best we can do (really)?

what drives differences between healthy individuals?

genomiccytometry.com

Van der Velde et al (2018), Swertz et al (2010) or Swertz & Jansen (2007)

are baseline immunological parameters correlated?

what is the impact of age/sex/BMI?

(how similar are healthy immune systems?)

are baseline immunological parameters in healthy individuals correlated?

age

count

bmi

PBMC

+

macrophages

+

whole blood

virus

TLR

non-mic

fungi

bacteria

n = 500

PBMC

mac

whole blood

Ig

mod

virus

TLR

non-mic

fungi

bacteria

TLR

fungi

bacteria

TLR

non-mic

fungi

bacteria

cytokines

n = 500

IgG

IgA

IgM

IgG1

IgG2

IgG3

IgG4

IL-18BPx

Resistin

Leptin

Adiponectin

AAT

IL-18

IL-1RA

Vitamin D

IL-1b

IL-6

VEGF A

CV

healthy people are <5% CV

PBMC

mac

whole blood

Ig

mod

virus

TLR

non-mic

fungi

bacteria

TLR

fungi

bacteria

TLR

non-mic

fungi

bacteria

cytokines

cytokines: 8 PCs >60% of variance

Spearman's rank correlations

are baseline immunological parameters in healthy individuals correlated?

pbmc mac whole blood

bacteria fungi TLR non-mic virus

UMAP

network analysis (a.k.a. graph theory)

undirected

directed

weighted

adjacency matrix

Classification (nodes)
- sequence-tagged antibody dataset
- classyDL

2. Interaction list (edges)

1,800 known, literature-supported interactions receptor- ligand interactions

Ramilowski et al., 2015

Mi et. al. 2019

Krzywinski, M. et al. 2009

1894 interactions

3. Calculate interaction scores (weight)

= \frac{1}{n_{celltype1}} \sum\limits_{i\epsilon \; cell type1} e_{i,receptor} \times \frac{1}{n_{celltype2}} \sum\limits_{i\epsilon \; cell type2} e_{i,ligand}

interaction score

e_{i,j} =

expression of gene j in cell i

n_{c} =

number of cells of cell type c

Kumar et al. Cell Reports 2018

(receptor, ligand,

celltype1, celltype2)

4. Network Analysis

measure	what	biological
dominant pairwise interactions	intrxn score > 0.005	dominant interactions
degree	# of connections to other nodes	central players which mediate information or distributed?
harmonic closeness centrality	degree to which nodes stand between each other (shortest paths)	most influential nodes
betweenness centrality	chokepoints of information transfer in networks	find a weakly connected cell type who still indispensable to certain transactions
authority	how valuable the information stored in that node	information quality of cell type (dx
hub	quality of nodes links	haves/have-nots

T8 Temra

Basophils

ClassicalMO

mDC

pDC

Th2

Th2

Basophils

ClassicalMO

mDC

pDC

Temra

strongest

pairwise

interaction threshold 0.00001

254/600 possible

degree

(connectivity)

harmonic

closeness

(influencers)

lo hi

between-

ness

(chokepoints)

interaction threshold 0.00001

254/600 possible

4. Network Analysis

measure	what	biological
dominant pairwise interactions	intrxn score > 0.005	dominant interactions
degree	# of connections to other nodes	central players which mediate information or distributed?
harmonic closeness centrality	degree to which nodes stand between each other (shortest paths)	most influential nodes
betweenness centrality	chokepoints of information transfer in networks	find a weakly connected cell type who still indispensable to certain transactions
authority	how valuable the information stored in that node	information quality of cell type (dx)
hub	quality of nodes links	haves/have-nots

authority

(info quality)

hub

(quality of connections)

lo hi

the humble basophil = best dx

interaction threshold 0.00001

254/600 possible

melanoma immune network

Tirosh & Izar et. al. Science 2016

data:

4645 single cells from 19 patients

includes malignant, stromal, immune, endothelial

cell classification:

k-finder

k-means

GSEA

I/O response rate <30%

combination I/O + need for dx

Turgeon et. al. (2018)

PBMC gene lists from genomic cytometry.com

melanoma

strongest interactions are 50X healthy

healthy

intrxn score

Text

v.

interaction threshold 0.01

33/64 possible

Macrophages

CAF cells

Endothelial

Macrophages (!)

Monocytes

NK cells

strongest

pairwise

Endothelial

NK cells

T cells

Macrophages

lo hi

cells with highest authority =

interacting with malignant

Text

1257 cells

R-L	role?
PSAP-CD1B	reported overepression in mel
PTMA-VIPR1	vasoactive intestinal peptide
GSTP1-TRAF2	biomarker for
LGALS3BP-ITGB1	integin, necessary for vasculogenic mimicry
HSP90AA1-CFTR	tumor aggression