ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2020

Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1964

Image and Data Analysis forSpatially Resolved Transcriptomics

Decrypting fine-scale spatial heterogeneity of tissue'smolecular architecture

GABRIELE PARTEL

ISSN 1651-6214ISBN 978-91-513-1003-9urn:nbn:se:uu:diva-419173

Dissertation presented at Uppsala University to be publicly examined in Room 4307,Ångströmlaboratoriet, Lägerhyddsvägen 2, Uppsala, Thursday, 29 October 2020 at 14:00 forthe degree of Doctor of Philosophy. The examination will be conducted in English. Facultyexaminer: Professor Roland Eils (Charité-Universitätsmedizin Berlin and Berlin Institute ofHealth).

AbstractPartel, G. 2020. Image and Data Analysis for Spatially Resolved Transcriptomics. Decryptingfine-scale spatial heterogeneity of tissue's molecular architecture. Digital ComprehensiveSummaries of Uppsala Dissertations from the Faculty of Science and Technology 1964. 59 pp.Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-513-1003-9.

Our understanding of the biological complexity in multicellular organisms has progressed attremendous pace in the last century and even more in the last decades with the advent ofsequencing technologies that make it possible to interrogate the genome and transcriptome ofindividual cells. It is now possible to even spatially profile the transcriptomic landscape oftissue architectures to study the molecular organization of tissue heterogeneity at subcellularresolution. Newly developed spatially resolved transcriptomic techniques are producing largeamounts of high-dimensional image data with increasing throughput, that need to be processedand analysed for extracting biological relevant information that has the potential to lead to newknowledge and discoveries. The work included in this thesis aims to provide image and dataanalysis tools for serving this new developing field of spatially resolved transcriptomics tofulfill its purpose. First, an image analysis workflow is presented for processing and analysingimages acquired with in situ sequencing protocols, aiming to extract and decode molecularfeatures that map the spatial transcriptomic landscape in tissue sections. This thesis alsopresents computational methods to explore and analyse the decoded spatial gene expressionfor studying the spatial molecular heterogeneity of tissue architectures at different scales.In one case, it is demonstrated how dimensionality reduction and clustering of the decodedgene expression spatial profiles can be exploited and used to identify reproducible spatialcompartments corresponding to know anatomical regions across mouse brain sections fromdifferent individuals. And lastly, this thesis presents an unsupervised computational method thatleverages advanced deep learning techniques on graphs to model the spatial gene expressionat cellular and subcellular resolution. It provides a low dimensional representation of spatialorganization and interaction, finding functional units that in many cases correspond to differentcell types in the local tissue environment, without the need for cell segmentation.

Keywords: iss, image, processing, clustering, deep learning, GCN, graph

Gabriele Partel, Department of Information Technology, Division of Visual Information andInteraction, Box 337, Uppsala University, SE-751 05 Uppsala, Sweden.

© Gabriele Partel 2020

ISSN 1651-6214ISBN 978-91-513-1003-9urn:nbn:se:uu:diva-419173 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-419173)

Defence on Zoom: https://uu-se.zoom.us/j/62337809210

detector

emission lter

exitation lterobjective

specimen

Dichromaticmirror

d

d =λ

2 ·NA,

λ NA

Hyb round 1 Hyb round 2 Hyb round 3 Coding scheme

osmFISH

mRNA 1

mRNA 2

mRNA 3

round

1

round

2

round

3

gene 1gene 2gene 3

round

1

round

2

round

3

gene 1gene 2gene 3gene n

MERFISH

mRNA mRNA mRNA

gene 1gene 2gene 3gene n

bitround 1 round 2 round 3

SeqFISH

mRNA mRNAmRNA

mRNA mRNA mRNA

SeqFISH (smHCR)

ln

n = 4l

Anchorprimer

BarcodeLinker(5th base)

Linker5' arm 3' arm

Padlock probe

mRNA

cDNA

cDNA

Cycle 4

Cycle 1

Cycle 2

Ligation& RCA

Padlock probeHybridization

Sample preparationmRNA

Reversetranscription& RNase H

RT primer

GACT

Sequencing

CTGAGNNN

CTGANANN

CTGANNNT

Cycle 3

CTGANNCN

41

42

43

44

Detectabletargets

DAPI +color channels

DAPI +decoded signals

Nucleichannel

Anchorchannel

Tchannel

Gchannel

Cchannel

Acahnnel

Cycle 1 Cycle 2 Cycle 3 Cycle 4 5th base

Mouse braina

b

c

d

IM IRT θ

L IM IR

θL(T (IM , θ), IR)

IR T (IM )

Nuclei image Gaussian blurring( = 5)

Watershedlabels

Inverted smoothed image

Thresholded image Histogram

Otzu threshold

Inverteddistance transform

Watershedlabels

ISE

Tw(I) = I − I ◦ SE

Originalimage

Top-hattransformed

◦

G = (V,E)V E eu,v ∈ E u v

wu,v

eu,v wu,v

st (G,w, s, t)

s− t C = (S, T ) Vs ∈ S t ∈ T

∑eu,v∈XC

wu,v

XC ST

XC = {eu,v ∈ E : u ∈ S, v ∈ T}.

s − t

π s tπ π

G = (V,E,w, s, t) f Gπ G

π s tfu,v < wu,v eu,v π

S T

cu,v ∈ R

f

c(f) =∑

eu,v∈Efu,v · cu,v.

π st

cu,v ππ π

Image Segmented image

source

sink

s

t

source

sink

u

v

s

t

wu,v

optimal s-tcut

K hf̂ X1, X2, . . . , Xn

f̂h(x) =1

nh

n∑i=1

K

(x−Xi

h

),

∫K(t)dt = 1 f̂h(x)

pq

DKL(p||q) =N∑i=1

p(xi) ·p(xi)

q(xi).

p qp q

y

y = b+∑i

xiwi,

xi wi b

f(y) = (y, 0)

n

softmax(y) =eyn∑i e

yi

Input layer (L0)

Hidden layer (L1)

Hidden layer (L2)

Output layer (L3)

j

z

Forward pass

i

k

l l

k

j

i

Backward passz

a

x1

x2

x3

w1w2

w3

Neuron Activationfunction

b

y f(y)

bKernel

Receptive eldActivation map

Convolutional layerc

3 ×3

−M∑c=1

yc (pc),

M ycc pc

c

lwij

∂l

∂wij=

∂l

∂z

∂z

∂yj

∂yj∂wij

,

lz z

jth yj yji j wij

M

MT cov(X)M

M cov(X)X

cov(X)M = λM,

λ d XX

YY = XM.

X

W HX ≈ WH.

X ∈ Rp×n

W ∈ Rp×r H ∈ R

r×n

p r p < r W

HW

XW W H

W,H‖X −WH‖2F , s.t.W,H ≥ 0.

XD d d < D d � D

X

k

G = (V,X,A) V ∈ Rn

A ∈ Rn×n

X ∈ Rm×n

GCNCNN

uK

uK

u

h0u = Xu ;

hku = σ

⎛⎜⎝Wk

∑v∈N (u)

hk−1v

|N (u)| +Bkhk−1u

⎞⎟⎠ , ∀k ∈ 1, . . . ,K ;

zu = hKu ;

Xu u h

N (u) u σ Wk Bk

zuu u

D

B

C

FE

A

D

B

C BEF

C

A

A

A

A

h0h1

z

a b

F

C E

C FB E FA

D

A

BC DB E FA

CA

BC D

B C

C E

F E

C F

E

C EB E FA

FC

c

zA

ec c ξ

ξ =K2

c

2m,

Kc

c mη

η =1

2m

∑c

(ec − γξ),

γ

a

sa ta

sb tb

b

Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1964

Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science andTechnology, Uppsala University, is usually a summary of anumber of papers. A few copies of the complete dissertationare kept at major Swedish research libraries, while thesummary alone is distributed internationally throughthe series Digital Comprehensive Summaries of UppsalaDissertations from the Faculty of Science and Technology.(Prior to January, 2005, the series was published under thetitle “Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology”.)

Distribution: publications.uu.seurn:nbn:se:uu:diva-419173

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2020