Integrative precision medicine: Local Data, Global Context Sven - - PowerPoint PPT Presentation

▶
integrative precision medicine local data global context
SMART_READER_LITE
LIVE PREVIEW

Integrative precision medicine: Local Data, Global Context Sven - - PowerPoint PPT Presentation

Aug 27, 2022 326 likes •445 views

Integrative precision medicine: Local Data, Global Context Sven Nelander Associate Professor Dept of Immunology, Genetics and Pathology Uppsala University Two fundamental challenges for cancer research Drug X! Drug X? Drug X??

slide-1
SLIDE 1

Integrative precision medicine: Local Data, Global Context

Sven Nelander Associate Professor Dept of Immunology, Genetics and Pathology
 Uppsala University


 


slide-2
SLIDE 2

Two fundamental challenges for cancer research

Drug X!

Global background data (n>100,000) Local cohort / patient data (n<1000)

TCGA ICGC L1000 100k genomes LINCS U-CAN HGCC SUS BILS KS UAS SCAN-B

Drug X? Drug X??

slide-3
SLIDE 3

Examples of global data: The Cancer Genome Atlas and LINCS/L1000

drug

1000 2000 3000 4000 5000 6000 7000

transcripts

500 1000 1500

slide-4
SLIDE 4

Example of local data: differential drug responses in Swedish brain tumors

D

Drug Cluster

YS_IN_CANCER 4e−05 GAP_JUNCTION 1e−05

< 2.22e−16 < 2.22e−16

Targets

Drugs Cell lines

AUC score 1

i ii

Associations

4.2e−11 4.2e−11 6.1e−09 5.2e−07 1.1e−05 3.4e−05 3.6e−05

Signature drugs

Subtype Cluster

MGMTmeth subtype sex HTR2A DRD2 TUBA1A TUBB4B TUBB TOP2B PSMB5

Kruskal−Wallis p: 0.49

0.001 0.01 0.1 1 10 100 0.0 0.5 1.0 1.5 Mean viabiliy (Carfilzomib) Dose (uM) 0.001 0.01 0.1 1 10 100 0.0 0.5 1.0 1.5 Mean viabiliy (Ixazomib) Dose (uM) 0.001 0.01 0.1 1 10 100 0.0 0.5 1.0 1.5 Mean viabiliy (Oprozomib) Dose (uM) 0.001 0.01 0.1 1 10 100 0.0 0.5 1.0 1.5 Mean viabiliy (MG-132) Dose (uM)

U3065_95 U3179 U3291 U3279 U3289 U3299 U3009 U3029 U3129 U3039 U3019

U3179 U3179 U3179 U3179 U3179 U3179

MG−132 W man: 0.079 p: 0.57986 Carfilzomib MG-132 Bortezomib Oprozomib

slide-5
SLIDE 5
  • 1. To develop computational methods for integrative modeling of big

cancer data sets

  • 2. To integrate small-cohort data and global modeling for target discovery

in cells from patients

  • 3. To make computational tools available as a web portal and efficient

standalone software

Sven Nelander Cancer systems biology Uppsala University Rebecka Jörnsten Mathematical statistics Chalmers Technical U Björn Nilsson Systems Medicine Lund University Erik Sonnhammer Bioinformatics Stockholm University George Michailidis Machine Learning University of Florida Terry Speed Biostatistics Walter and Eliza Hall Institute
 UC Berkeley

Goals and team

slide-6
SLIDE 6

Project organisation

Rebecka Jörnsten (co-lead PI) Jonathan Kallus (PhD student) Szilard Nemes (postdoc) ++ New PhD student ++ Erik Sonnhammer (co-PI) Daniel Morgan (PhD student) Deniz Secilmis (PhD student) ++ New postdoc ++ Sven Nelander (lead PI) Patrik Johansson (PhD student) Emil Rosén (PhD student) ++ Caroline Wärn (bioinformatician) ++ Anders Sundström (bioinformatician) ++ Cecilia Krona (researcher) Elin Almstedt (PhD student) New Postdoc ++ Björn Nilsson (co-PI) Ludvig Ekdal (PhD student) Maroulio Pertesi (researcher) Ram Ajore (researcher) Terry Speed (collaborator) Ingrid Lönnstedt (reseacher) George Michailidis (collaborator) Postdoc

Analysis of response data (2.1) Cell response RNA-seq (2.2) Data integration methods (1.1,1.2) Deep models (1.1) Shared Resources (GitHub, server, AWS) Shared Resources (molecular biology)

Color Key: Computational biologist Mathematician Experimental systems biologist ++ = recruited as part of SSF program implementation

slide-7
SLIDE 7

Aim 1: new data integration methods

Jörnsten et al, MSB 2011; Kling et al, NAR 2015; Kling et al EBioMedicine 2016; 
 Schmidt et al, OncoTarget 2016; Yiang et al, Cell Reports 2017

A

shared co-variation across cohorts

data modalities

Uppsala cohort TCGA cohort Individual patient

shared co-variation across data modalities

slide-8
SLIDE 8

Aim 2: demonstration study on brain tumor cells from Swedish patients

drug

1000 2000 3000 4000 5000 6000 7000

transcripts

500 1000 1500

L1000/LINCS (extreme scale profiling of drugs in 77 cell lines)

Drug response profiles
 (collected at UU,LU by our team)

drug

1000 2000 3000 4000 5000 6000 7000

transcripts

500 1000 1500

Machine learning models Calibration study Use of models to predict interventions

slide-9
SLIDE 9

MATCH (n1)-[e00:HGCC]-(n0), (n0)-[e2]-(n3), (n3)-[e4]-(n5) WHERE n0:isTF AND n0.type="expr" AND n1.type="drug" AND n3.type="cna" AND n5.type="clinical" WITH (e00.weight+e2.weight+e4.weight) as score, n0, n1, n3, n5, e00, e2, e4 RETURN score, n0, n1, n3, n5, e00, e2, e4 ORDER BY score DESC

Interactive motif drawing Ranking and relation of matching motifs

Aim 3: new tools for cancer data mining

Patrik Johansson

1 2 3

Johansson et al, in preparation

slide-10
SLIDE 10

Current status and plans onwards

  • 3 workshops held + 1 international meeting hosted (SCAS)
  • 1 project paper accepted, 2 in preparation
  • Research visits Uppsala <-> Gothenburg and Lund
  • Discussions with SciLifeLab leadership on next-generation data integration
  • Total team grown by 6 new members (UU, SU, CTH). 2 recruitments to go
AKT1 EGFR ERBB2 MAPK8 MTOR TP53 MAPK1 MAPK3 CASP8 HIF1A JUN TUBB4B BCL2 CASP2 CASP7 CDK2 CFLAR GSTP1 HDAC1 PARP1 VEGFA CASP3 ABCB1 ABCG2 CYP3A4 ERBB3 IGF1R JAK2 MCL1 MMP9 SRC
  • 3. Aggregated scores of

small molecules

  • 4. Enriched protein

targets

  • 2. RNA response profiles
  • f >7000 drugs in 77 cell lines
  • 1. RNA and clinical data

from multiple cohorts

target analysis

Stage MYCN Hazard ALK 1 chemical compound score FDR=5% FDR=10% FDR=20% FDR=30% Stage MYCN Hazard ALK

hits

drug 1000 2000 3000 4000 5000 6000 7000 transcripts 500 1000 1500 Stage MYCN Hazard ALK

aggregate scoring across cohorts and cell models

p=10-10 p=10-20

Big data processing pipeline

1 2 3

Lönnstedt et al, SAGMB 2017 Almstedt et al, in preparation Johansson et al, in preparation