Transcript PPT - NIH LINCS Program

GATCACTGGCATGCATCGATCGACTGACTGCGGCATGCGCG
ATCGACTGGCGATCAAACAGTCACGCGCATCGATCGACTGA
GATCGCGGCATCGCGACGCGGATAAATACGAGCACTACAAA
TGACTACGGGATTTTACGCGCGATACGACTGACTGACTAGC
GATCACTGGCATGCATCGATCGACTGACTGCGGCATGCGCG
LINCS Fall Consortia Meeting
ATCGACTGGCGATCAAACAGTCACGCGCATCGATCGACTGA
GATCGCGGCATCGCGACGCGGATAAATACGAGCACTACAAA
Broad Institute U54 Team
TGACTACGGGATTTTACGCGCGATACGACTGACTGACTAGC
TGACGATCGAGAGACTCG01010001010101000101010101001
Todd Golub, co-PI
0010101010100000011110101111101001010101000111011101
Wendy Winckler, co-PI
0111101101010111001010101000111010101001100101110101
Aravind Subramanian, Team Leader
0111010010001010100011110101000010101010100010100011
0010101000101011110101000100100100101010001000001011
October 27, 2011
0010101010100000011110101111101001010101000111011101
0111101101010111001010101000111010101001100101110101
0111010010001010100011110101000010101010100010100011
0010101000101011110101000100100100101010001000001011
0101001010000101111101001010010101011101010010101001
BASIC
DISCOVERIES
GENETIC
CONNECTIONS
THERAPEUTIC
IMPACT
PATHWAYS
DISEASE STATES
TOOL COMPOUNDS
DRUGS
GWAS
TCGA
RNAi
CHEMICAL
SCREENS
NAT’L PRODUCTS
SLOW (SOME NEVER START)
DOES NOT SCALE
NO LEVERAGE
DIAGNOSTICS
LINCS as a Solution
• perturbations scalable to genome
• high information content read-outs
(e.g. gene expression)
• inexpensive
• mechanism to query database
Toward a reduced representation of the transcriptom
gene expression is correlated
genes
samples
Reduced Representation of Transcriptome
reduced
representation
transcriptome
genome-wide
expression profile
computational
inference model
‘landmarks’
100
60
40
20
A. Subramanian, R. Narayan
100
300
500
700
1000
1000
1500
2000
5000
10000
14812
0
22283
~ 100,000 profiles
% connections
80
80%
number of landmarks measured
1000-plex Luminex bead profiling
AAAA 3'
5'
 RT
3'
5'
5'-PO4
|
TTTT
3'
Luminex Beads
(500 colors,
2 genes/color)
 ligation
5'
5'
 PCR
 hybridization
001
Reagent cost:
$3/sample
Validation of L1000 approach
Gene-level validation
12
1000-plex-Luminex
11
10
9
92% R2 > 0.6
Similar to AFFX vs ILMN
8
7
6
5
4
6
8
10
12
14
Affymetrix
Affymetrix ($500)
C-Map Connections
Affymetrix
simulation
Luminex ($5)
1,000-plex
Connections
Published (32)
Internal (152)
26 (80%)
121 (80%)
28 (86%)
142 (94%)
Putting it all together
Illustration: Bang Wo
Cell Types
GTEx
Primary hTERT-immortalized
cells
Patient-derived iPS cells*
Banked primary cells* (T-cells,
macrophages, hepatocytes,
myocytes, adipocytes)
Cancer cell lines
* in assay optimization
2-3 weeks
Cell Repository
(e.g. Coriell)
3-4 weeks
4-6 weeks
Reprogramming
[Oct4, Sox2, Klf4, Myc]
Neural Differentiation
Astrocyte
somatic cell
isolation
fibroblasts
Oligodendrocyte
Neural
progenitors
Neuron
Perturbagens
Small-molecules (n=4,000)
Genes (n=3,000)
Automated Quality Control Measures
Overall failure rate ~ 8%
LINCS Proposal (~ 600,000 profiles)
4,000 compounds
• 1,300 off-patent FDA-approved drugs
• 700 bioactive tool compounds
• 2,000 screening hits (MLPCN + others)
2,000 genes (shRNA + cDNA)
• known targets of FDA-approved drugs (n=150)
• drug-target pathway members (n=750)
• candidate disease genes (n=600)
• community nominations (n=500)
20 cell lines
• emphasis on reproducibility and availability
• cancer and primary, non-cancer
• some ‘doubling down’ to assess intra-lineage diversity
Progress to date
http://www.broadinstitute.org/lincs_beta/
DATA RELEASE (BETA)
proposed
actual
projected
Signature of p53 ORF
p53 vs. empty vector
• p53 is NOT a Landmark Gene
• p53 pathway is #1 pathway of 512 in MSigDB
P < 0.001
Ramnik Xavier
Making connections in primary macrophages
NF-kB pathway genes
(all INFERRED)
pathway rank: 1/512
LPS
pathways
curated from
literature
(n=512)
Jens Lohr
Prioritizing human genetics candidates
Ramnik Xavier, MGH
Signatures of genetic variants connect to disease genes
Ramnik Xavier, MG
Disease variants connect to pathways
e.g. CD40 to ATG16L1 (both regulators of autophagy)
Ramnik Xavier, MGH
ERG transcription factor
important in hematopoietic stem cells, prostate cancer
ERG-BINDING SMALL-MOLECULES
Defining a gene expression signature of ERG activity
integrating experimental and clinical data
Gain of Function:
Primary prostate + hTERT
+ST +AR +/-ERG
Loss of Function:
VCaP cells +/- ERG shRNA
120
Patient Samples:
Physician’s Health Study
3/69 ERG-binders inhibit ERG gene expression program
L1000 as primary small-molecule screen read-out
12,985 compounds screened for ERG signature
Name
Rank
Library
BRD-K42581894-001-01-1
1
DOS
BRD-K42581894-001-01-1
2
DOS
BRD-K14408783-001-01-5
3
DOS
Wortmannin
4
Bioactives
BRD K78122587
5
ChemDiv
BRD-K91899208-001-01-8
6
DOS
BRD-K24750847-001-01-2
7
DOS
BRD-K18273607-001-02-1
8
DOS
BRD-K76892938-001-01-9
9
DOS
AZD2281 (Olaparib)
10
Bioactives
BRD-K86715531-001-01-1
11
DOS
BRD-K95688283-001-01-9
12
DOS
BRD-K99179945-001-01-5
13
DOS
Analytical and software challenges
1.
2.
3.
4.
5.
Infrastructure: data and compute server
Optimization of connectivity metrics and statistics
Optimization of inference models (context-aware)
UI: query tools and results visualization
Addressing off-target effects of perturbagens
Aravind Subramanian
Wendy Winckler
Justin Lamb
Computational
Rajiv Narayan
Josh Gould
RNAi Platform
Laboratory
Chemical Biology Platform
Dave Peck
Genetic Analysis Platform
Willis Reed-Button
Broad Program Scientists
Xiaodong Lu