diego di bernardo telethon institute of genetics and
TRANSCRIPT
![Page 1: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/1.jpg)
Gene network inference in diseases and drug discovery
Diego di BernardoTelethon Institute of Genetics and Medicine
19 Dicembre 2007, Napoli
![Page 2: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/2.jpg)
dibernardo.tigem.it
Gene Networks
cell membrane
metabolites
proteins
RNA
genes
transcriptnetworks
proteinnetworks
metabolicnetworks
Our focus: methods to decode transcription regulation networks
![Page 3: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/3.jpg)
dibernardo.tigem.it
Reverse engineering (or system id) gene networks:
?Unknown network Inferred network
![Page 4: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/4.jpg)
dibernardo.tigem.it
Aim of reverse engineering: gene function and drug MOA
Experiments can be performed as a time-series or at steady-state
D di Bernardo et al, Nature Biotechnolgy, 2005; TS Gardner, D di Bernardo et al, Science, 2003;
![Page 5: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/5.jpg)
dibernardo.tigem.it
Reverse-engineering strategy:
• Choose a model• Choose a fit criterion (cost function) to measure the fit of the model to the data• Define a strategy to find the parameters that best fit the data (i.e. that minimise
cost function)• Perform appropriate experiments to collect the experimental data:
![Page 6: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/6.jpg)
dibernardo.tigem.it
Comparision of different models
Mukesh Bansal, Vincenzo Belcastro, Alberto Ambesi, Diego di Bernardo. How to infer gene networks from expression profiles. Molecular Systems Biology, 2007.
Bayesian network (Hartemink, A.(2005) Nature Biotechnology, 2005.)
Information-theoretic: (Basso et al.,Nature Genetics, 2006)
ODEs (Gardner, di Bernardo et al,Science, 2003; di Bernardo et al,Nature Biotechnolgy, 2005)
![Page 7: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/7.jpg)
dibernardo.tigem.it
Understanding gene function in a genetic disease:
P63 is a transcription factor primarily expressed in epithelia including theproliferative compartments of the skin.
It plays an essential role in modulating cellular differentiation by unknownmechanisms.
Mutation in its DNA binding and SAM domains have been linked to fivehuman malformation syndromes.
![Page 8: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/8.jpg)
dibernardo.tigem.it
•Lack of epidermis and hair follicles•Defects in all squamous epithelia•Limb truncations•Craniofacial defects: lack of tooth primordiaand eyelids•Maxilla and mandible are truncated andsecondary palate fails to close
p63-deficient mice
•P63 is a Transcription Factor (unknowntargets)•p63 gene has three different promoter thatgives rise to two different transcripts (TA and∆N);•each transcript has three alternative splicingisoforms (α,β,γ)
The transcription factor p63:
![Page 9: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/9.jpg)
dibernardo.tigem.it
Model structure: Ordinary differential equations
dX1/dt = a2 X2 + a6 X6 + a9 X9 + a12 X12
promoters
RNAs
Directed graph
+b1u
u
p63
![Page 10: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/10.jpg)
dibernardo.tigem.it
x’1(t) = a11x1(t) +a12x2(t) +...+a1nxn(t) + b1u(t)
........................................
x’n(t) = an1x1(t) +an2x2(t) +...+annxn (t) + bnu(t)
Gene effectunknown
Model structure:
![Page 11: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/11.jpg)
dibernardo.tigem.it
x1(tk+1) = ad11x1(tk) +ad12x2(tk) +...+ad1nxn(tk) + bd1u(tk)
........................................
xn(tk+1) = adn1x1(tk) +adn2x2(tk) +...+adnnxn (tk) + bdnu(tk)
Or in matrix format:
x(tk+1) =Adx(tk)+bdu(tk)For k=1..M
X(tk+1)=AdX(tk)+bdu(tk)Where (NxM)=(NxN)(NxM)+(Nx1)(1xM)
Model structure - Discrete model:
![Page 12: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/12.jpg)
dibernardo.tigem.it
Fit criterion (Least Squared Error):
NxM = Nx(N+1) (N+1)xM
![Page 13: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/13.jpg)
dibernardo.tigem.it
Solution search strategy:
• Usually N>>M (more unknown than data points; i.e. 1000 genes e 20 points)
• From the previous equation we know that a unique solution will exist only ifN≤M-1
• We need a trick to reduce number of unknowns by dimensional reduction:– Regression with Variable selection– Clustering– Regression with SVD
![Page 14: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/14.jpg)
dibernardo.tigem.it
Solution search strategy: Regression using PCS/SVD
• Singular Value Decomposition:– Y=UDV’ (N+1xM)=(N+1xM)(MxM)(MxM)
– Y=UrDrVr’ (N+1xM)=(N+1xK)(KxK)(KxM)
– X(tk+1) =HY= (NxN+1)(N+1xM) =
– X(tk+1) =HUrDrVr’=(HUr)DrVr’=(NxK)(KxK)(KxM)
![Page 15: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/15.jpg)
dibernardo.tigem.it
Preliminary Results on noisy “in silico” data
Network of 10 genes (A) Target prediction in 1000 gene network (B)
![Page 16: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/16.jpg)
dibernardo.tigem.it
Application to p63:
![Page 17: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/17.jpg)
dibernardo.tigem.it
Localization of ∆Np63 upon tamoxifen treatment by immunofluorescence
Murine keratinocytes are infected withGinco ErDNp63 retrovirus. The cells arecollected at the following time points fromTamoxifen induction: 1hr, 2hrs,4hrs, 8hrs.
Immunofluorescence plot: red andblue lines represent two differentimmunofluorescence experiments.
0hr 8hr
!
x(t) = A " x(t) + Bu(t).
![Page 18: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/18.jpg)
dibernardo.tigem.it
Hierarchical Clustering of 786 expression profiles:
![Page 19: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/19.jpg)
dibernardo.tigem.it
Application of TSNI on 786 expression profiles:
TSNI(B matrix)
![Page 20: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/20.jpg)
dibernardo.tigem.it
Validation of predicted targets: (1) ChIP-chip
• We selected the top 100 predicted targets and as control thebottom 200 predicted genes.
•For each gene we analysed 20 Kbp upstream+gene sequencevia ChIP-chip (Agilent 200k custom-array)
•We found about 300 p63-binding sites bu ChIP-chip with ap-value<0.01
![Page 21: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/21.jpg)
Validation of predicted targets: (1) ChIP-chip
![Page 22: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/22.jpg)
dibernardo.tigem.it 22
AB
Validation of predicted targets: (1) ChIP-chip
![Page 23: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/23.jpg)
dibernardo.tigem.it 23
TSNI results are confirmed by ChIP-chip
![Page 24: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/24.jpg)
dibernardo.tigem.it 24
RNA interference against the p63 DNA binding domain
Knocking down p63 expression via siRNA:
![Page 25: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/25.jpg)
dibernardo.tigem.it 25
A B
TSNI results are confirmed by siRNA
![Page 26: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/26.jpg)
dibernardo.tigem.it
Application of TSNI to drug discovery :
Drug treatment Drug targets
![Page 27: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/27.jpg)
dibernardo.tigem.it
A “steady-state” approach to drug target identification:
D di Bernardo et alNature Biotechnolgy, 2005
![Page 28: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/28.jpg)
dibernardo.tigem.it
MNI identifies target of itraconazole in S. cerevisiae
Expression Change MNI Predictions
Filter through MNI-inferred network model
Itraconazole treatment: a known target is ERG11
Erg11Erg11
![Page 29: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/29.jpg)
dibernardo.tigem.it
MNI results:
Pathway data from GO database
![Page 30: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/30.jpg)
dibernardo.tigem.it
MNI identifies MOA for 8 out of 9 drugs:
![Page 31: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/31.jpg)
+
Acknowledgements (http://dibernardo.tigem.it for more info):
…and Dr Caterina Missero (CEINGE, Napoli) for supervision ofexperiments; Prof. Tim Gardner and Prof. Jim Collins (BostonUniversity, USA) for MNI/NIR related work.
Dr Alberto Ambesi
(Sequence analysis)
Giusy Della Gatta
(Experimental work)
Mukesh Bansal
(Computational work)
Velia SicilianoLucia MarucciGiulia Cuccato
Francesco Iorio
Vincenzo Belcastro
Mario Lauria
![Page 32: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/32.jpg)
dibernardo.tigem.it
•First filter• Area under curve
•Second filter• Standard deviation is computed at each time pointusing smoothing and generalized cross validationalgorithm
• Statistical test for significance (Chi-Square)
Selection of Significant Genes in Time Series experiment:
![Page 33: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/33.jpg)
dibernardo.tigem.it
in silico with dual perturbation
dX1/dt = a2 X2 + a6 X6 + a9 X9 + a12 X12+b11u1 + b12u2
p63 Tamoxifen
![Page 34: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/34.jpg)
dibernardo.tigem.it
Improvement of TSNI to identify the network (A)
•Integral Approach instead of differential approach
Mukesh Bansal, Diego di BernardoInference of gene networks from temporal gene expression profiles. IET (Under Review)
•Bootstrapping
![Page 35: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/35.jpg)
dibernardo.tigem.it
(1a) Time Series Network Identification Algorithm (NEW) :
P63 directly regulates at early times AP1 complex and markers of terminal keratinocytedifferentiation -Manuscript submitted-
![Page 36: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/36.jpg)
dibernardo.tigem.it
Identified novel anticancer compound via chemical screen
• PTSB inhibits growth in yeast and tumor cell lines
In collaboration with Schaus and Elliot laboratoriesDept. of Chemistry, Boston UniversityCenter for Methodology and Library Development (CMLD), Boston U.
![Page 37: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/37.jpg)
dibernardo.tigem.it
MNI identifies two enzyme targets of PTSB
Identifies thioredoxin (TRX2) and thioredoxin reductase (TRR1)
![Page 38: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/38.jpg)
dibernardo.tigem.it
TRR1/TRX2 activity inhibited in presence of PTSB
Assay:
Thioredoxin reduction of dithio(bis)nitrobenzoic acid (DTNB)
• Product of reaction = thiolate anion, measured via A412
0 uM PTSB
1 uM PTSB
5 uM PTSB
50 uM PTSB
![Page 39: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/39.jpg)
dibernardo.tigem.it
MNI algorithm overview and application to S. cerevisiae
4) Select top 50 genes, i.e. top 50 predicted drug targets
3) Rank genes using model learned in step 1
2) Measure profile in response to drug 15 compoundstested
1) Learn Model 6000 genes 512 microarrays
5) Find common pathway among top 50 predicted targets
![Page 40: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/40.jpg)
dibernardo.tigem.it
Geno Ontology Filtering
![Page 41: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/41.jpg)
dibernardo.tigem.it
?
MNI: Microarray Network Identification
![Page 42: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/42.jpg)
dibernardo.tigem.it
x’1(t) = a11x1+a12x2+...+a1nxn + b1u
........................................
x’n(t) = an1x1+an2x2+...+annxn + bnu
Or in matrix format:
x’=Ax+bu
Drug effect unknown
Model structure:
![Page 43: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/43.jpg)
dibernardo.tigem.it
0 = a11x11+a12x21+...+a1nxn1 + b1u
........................................
0 = a11x1h+a12x2h+...+a1nxnh + b1u
Fit criterion and search solution strategy – phase 1:
Choose only the h experiments where gene 1 has not been perturbed
and solve the eqs. for gene 1 with a11≠0.
Repeat for all the genes and obtain matrix A
…we still did not say how to find the h experiments where gene ihas not been perturbed, this is done simply by choosing those experiments where the gene has changed less…
![Page 44: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/44.jpg)
dibernardo.tigem.it
Fit criterion and search solution strategy – phase II:
bu=-Axd
Say xd the expression profile following the drug treatment, thenthe predicted targets of the drug are:
![Page 45: Diego di Bernardo Telethon Institute of Genetics and](https://reader030.vdocuments.net/reader030/viewer/2022012711/61aa8bcdf9bbaf11e87429ac/html5/thumbnails/45.jpg)
dibernardo.tigem.it
Other projects going on in our lab:
- REVERSE ENGINEERING BASED ON MI AND BAYESIAN APPROACH- IDENTIFICATION OF DRUG TARGETS- SYNTHETIC BIOLOGY - COBIOS (coordinator) 120kE/yr for 3 yrs for my lab