Supplementary material: Linking Signaling Pathways to Transcriptional Programs in Breast Cancer



Download 0.63 Mb.
Page5/5
Date conversion08.07.2018
Size0.63 Mb.
1   2   3   4   5

Supplementary Tables



Table S1. TF subtype specificity assessed by TF mRNA expression level vs inferred TF activity.














TF

p-value TF-mRNA

p-value TF-activity

Subtype specificity

HMX3-HMX2

8.3E-01

8.7E-11

HER2

TTF-1

3.6E-01

5.3E-15

LumA

FXR-FXR2

3.4E-01

1.9E-14

LumA/B

MECOM

3.3E-01

1.6E-22

LumA/B/HER2

FXR-FXR1

3.0E-01

1.9E-14

LumA/B

HMG-HMGN1

2.5E-01

1.2E-22

Basal-like

VSX2-VSX1

1.0E-01

3.6E-10

LumA

FOXF1

8.0E-02

4.9E-16

LumA/B

SMAD4

7.6E-02

2.6E-16

LumA

IRF2

3.2E-02

2.5E-19

LumA

HMG-HMGN2

3.1E-02

1.2E-22

Basal-like

GTF2I

2.4E-02

6.6E-23

LumA/B

MEIS1

1.3E-02

7.0E-29

LumA/B

NKX2-2

9.0E-03

1.0E-18

LumA

ZBTB14-ZFP161

3.9E-03

6.7E-11

HER2

HMG-HMGB1

1.9E-03

1.2E-22

Basal-like

HMG-HMGB3

2.9E-05

1.2E-22

Basal-like

HMG-HMGB2

3.1E-06

1.2E-22

Basal-like

HMG-HMGA3

2.0E-06

1.2E-22

Basal-like

HMG-HMGN3

2.0E-06

1.2E-22

Basal-like

FOXA1

1.9E-06

4.4E-15

LumA

NFATC4

3.5E-08

2.3E-22

Basal-like

SOX9

3.6E-09

2.8E-22

Basal-like

ETS1

1.9E-09

1.2E-22

Basal-like

CEBPB

3.2E-16

2.6E-23

Basal-like

GATA3

2.5E-24

1.1E-25

LumA/B

IRF10

NA

7.0E-25

LumA/B


Table S2: Signaling protein subtype specificity assessed through protein expression and inferred protein activity.














Protein

p-value protein expression

p-value protein activity

Subtype specifity

AKT1/AKT2/AKT3 (pS473)

7.9E-01

6.0E-11

HER2

RPS6 (pS240)

5.1E-01

1.3E-11

LumB

MAPK14 (pT180)

4.4E-01

4.6E-15

Basal-like

RPS6 (pS235)

4.1E-01

3.6E-18

HER2/LumB

STAT5A

3.5E-01

1.2E-15

Basal-like

FN1

3.2E-01

7.0E-17

LumA/B

RB1 (pS807)

2.9E-01

5.3E-14

Basal-like

CTNNB1

1.4E-01

1.5E-15

Basal-like

KDR

5.4E-02

1.1E-16

Basal-like

AKT1/AKT2/AKT3

1.1E-02

1.1E-18

HER2/LumA

PEA15

3.8E-04

2.1E-11

LumA

PTGS2

3.7E-04

4.5E-18

Basal-like

PDK1 (pS241)

1.4E-04

1.7E-15

LumA

IGFBP2

3.0E-05

4.7E-13

LumB

CAV1

1.4E-06

7.2E-19

LumA/B

CCND1

2.1E-07

3.3E-22

LumA/B

CHEK2

7.8E-09

3.7E-21

Basal-like

ERBB2

1.8E-09

5.7E-17

HER2

CCNB1

1.8E-09

4.4E-17

Basal-like

MSH2

1.7E-09

2.1E-22

Basal-like

ERBB2 (pY1248)

6.8E-11

3.6E-17

HER2

MSH6

3.4E-11

1.3E-20

Basal-like

CDH3

3.0E-11

1.0E-23

Basal-like

WWTR1 (pS89)

1.7E-12

5.3E-20

Basal-like

KIT

1.1E-12

2.2E-22

Basal-like

BCL2

1.2E-13

3.9E-15

LumA/B

PR

2.1E-14

1.2E-18

LumA

CCNE1

5.5E-15

1.0E-24

Basal-like

INPP4B

1.3E-16

2.2E-22

LumA/B

AR

1.1E-21

1.4E-23

HER2/LumA/B

GATA3

6.1E-22

3.8E-25

LumA/B

ESR1

7.9E-28

3.8E-20

LumA/B



Table S3. Rank correlation of inferred TF activity with measured protein expression and with inferred protein activity for the TFs in the RPPA data set.


TF

Protein

ρ

ρ

(TF activity vs. protein expression variation)

(TF activity vs inferred protein activity)

GATA3

GATA3

0.67

0.72

SMAD3

SMAD3

0.20

0.27

JUN

JUN (pS73)

0.16

0.32

SMAD4

SMAD4

0.16

0.34

FOXO3

FOXO3a

0.15

0.01

MYC

MYC

0.10

−0.08

FOXO3

FOXO3a (pS318_S321)

0.09

0.25

STAT5_STAT6

STAT5A

0.06

0.06

AR

AR

0.05

0.11

XBP1

XBP1

−0.01

−0.32

SMAD1

SMAD1

−0.03

−0.19

TP53

TP53

−0.10

−0.22

STAT3

STAT3 (pY705)

−0.13

−0.10

NFKB1

NFKB1 (p65_pS536)

−0.28

−0.50

ESR1

ESR1 (pS118)

−0.31

−0.10

ESR1

ESR1

−0.42

−0.34


Table S4. Drugs and their targets (order as in Figure 3A).


Compound

Target/Mechanism of action

Carboplatin

DNA cross-linker

Cisplatin

DNA cross-linker

Erlotinib

EGFR

Docetaxel

TUBB1, BCL2

TPCA−1

IKBKB

Lestaurtinib

FLT3, NTRK1

BIBW2992

EGFR, ERBB2

GSK1120212

MAP2K1, MAP2K2

BEZ235

PIK3CA

Vinorelbine

TUBB

Gefitinib

EGFR

Glycyl−H−1152

ROCK2

Topotecan

TOP1

Gemcitabine

Pyrimidine animetabolite

Sorafenib

KDR

XRP44X

ELK3

Bortezomib

PSMD2, PSMB1, PSMB5, PSMB2, PSMD1

AZD6244

MAP2K1, MAP2K2

5−FdUR

TYMS, DNA, RNA

PD173074

FGFR3

5−FU

TYMS, DNA, RNA

Triciribine

AKT, ZNF217

TCS JNK 5a

MAPK9, MAPK10

AG1478

EGFR

Oxaliplatin

DNA cross-linker

CPT−11

TOP1

ZM 447439

AURKA

ICRF−193

TOP2BA, TOP2AB

VX−680

AURKA, AURKB, AURKC

GSK1487371

PIK3CG

GSK1070916

AURKB, AURKC

Paclitaxel

TUBB1, BCL2

GSK923295

CENPE

GSK461364

PLK1

Ispinesib

Kinesin

Etoposide

TOP2A

Methotrexate

DHFR

Pemetrexed

TYMS, DHFR, GART

Ixabepilone

TUBB3

CGC−11047

Polyamine analogue

MLN4924

NAE1

TGX−221

PIK3CB

GSK2119563

PIK3CA

Rapamycin

MTOR

Lapatinib

EGFR, ERBB2

Temsirolimus

MTOR

17−AAG

HSP90AA1

Tamoxifen

ESR1



Table S5. Univariate survival analysis for luminal cohort (Luminal A + Luminal B) from METABRIC.

























Predicted Protein Activity

Gene expression profiles

Covariate

Coef(bi)

HR[exp(bi)]

p-value

Covariate

Coef(bi)

HR[exp(bi)]

p-value

ESR1

9.6

1.5E+04

1.7E-01

STAT5A

-0.7

0.5

6.4E-06

STAT5A

-87.8

7.5E-39

4.6E-08

CCNB1

0.6

1.8

1.7E-05

ERBB2 (pY1248)

119.3

6.2E+51

3.6E-07

CTNNA1

1.0

2.7

4.6E-05

SMAD4

-436.3

3.1E-190

3.5E-06

PGR

-0.3

0.7

2.0E-04

KIT

-47.8

1.7E-21

2.7E-06

NCOA3

1.2

3.2

1.9E-04

PGR

-39.6

6.5E-18

5.1E-06

BCL2

-0.7

0.5

4.0E-04

YWHAE

-424.1

6.7E-185

9.8E-06

KRAS

1.4

4.0

4.8E-04

TP53

110.1

6.4E+47

9.0E-06

RPS6

-0.8

0.5

6.6E-04

ERBB2

67.8

2.9E+29

1.6E-05

IRS1

-0.4

0.7

7.1E-04

COL6A1

-64.1

1.5E-28

2.9E-05

MAPT

-0.5

0.6

8.3E-04

COX2

-57.3

1.4E-25

3.6E-05

ANLN

0.5

1.7

1.7E-03

RPS6

107.9

7.5E+46

4.8E-05

ATM

-1.3

0.3

2.2E-03

EIF4EBP1

-157.8

3.0E-69

5.6E-05

SRC

1.1

3.1

2.2E-03

ACACA

64.4

9.1E+27

1.1E-04

PXN

0.9

2.5

2.5E-03

PARK7

-87.0

1.7E-38

1.6E-04

FN1

0.8

2.2

3.2E-03

RPS6 (pS235_S236)

33.4

3.2E+14

4.1E-04

NOTCH3

0.5

1.7

3.0E-03

RPS6 (pS240_S244)

35.3

2.2E+15

5.3E-04

DVL3

0.7

2.0

2.9E-03

CDKN1B

-111.4

4.0E-49

5.3E-04

EEF2

-0.6

0.5

3.3E-03

EGFR (pY1173)

-217.5

3.4E-95

5.5E-04

KIT

-0.3

0.7

4.2E-03

BAX

-105.3

1.9E-46

6.3E-04

PARK7

-0.7

0.5

6.4E-03

CCNB1

26.5

3.2E+11

6.8E-04

CDK1

0.4

1.5

9.3E-03


Table S6. Comparison of drug response prediction across data sets.

Drug response prediction models, trained on inferrred protein activities, generalized to the CCLE breast cancer cell lines for 5 out of 7 cases.




Drug

#cell lines Neve

#cell lines CCLE

Drug response correlation

(ρ: Neve-GI50,CCLE-AUC)


Drug response SD: Neve

Drug response SD: CCLE

Neve et al. Performance

Test (CCLE) Performance

(ρ)

(ρ)

17-AAG

32

26

0.56

0.69

1.12

0.59

0.15

AZD6244

24

26

0.39

0.63

0.69

0.52

0.56

Erlotinib

34

26

0.32

0.51

0.62

0.68

0.40

Lapatinib

30

26

0.78

0.67

1.07

0.67

0.39

Pactitaxel

32

26

0.44

0.67

1.28

0.38

0.46

Sorafenib

33

26

-0.49

0.62

0.29

0.66

-0.09

Topotecan

31

26

0.80

0.77

1.26

0.62

-0.17



Table S7. Summary of datasets used in the study.

Dataset

Summary

Reference

TF binding sites

Transcription factor (230) binding site defined in the TRANSFAC for each gene from MSigDB

MSigDB (Liberzon et al. 2011)

TCGA BRCA mRNA

532 tumor, 61 normal samples

TCGA (Cancer Genome Atlas Network 2012)

TCGA BRCA RPPA

397 tumor samples - 164 proteins and phosphoproteins,

TCGA (Cancer Genome Atlas Network 2012)

TCPA Breast cancer cell line RPPA

187 protein and proteins and phosphoproteins

Li et al. (Li et al. 2013)

Breast cancer cell line mRNA

51 breast cancer cell lines

Neve et al. (Neve et al. 2006)

Compound and cell line
screening data

Drug-response data for 31 cell lines treated with 74 drugs

Heiser et al. (Heiser et al. 2012)

Cancer Cell Line Encylopedia (CCLE)

Drug-response data for cell lines treated with 24 compounds and microarray mRNA

Barretina et al. (Barretina et al. 2012)

METABRIC BRCA cohort

~2000 tumors with survival with microarray mRNA data

Curtis et al. (Curtis et al. 2012)

TRANSBIG BRCA cohort

~198 tumors with survival and microarray mRNA data

Desmedt et al. (Desmedt et al. 2007)

NKI BRCA cohort

~337 tumors with survival and microarray mRNA data

van Vijver et al. (van de Vijver et al. 2002)



Supplementary Materials

Supplementary Material 1. TCGA patient list

List of training and test TCGA patients used in the study.



Supplementary Material 2. RPPA protein list

List of protein RPPA identifiers used in the study.



Supplementary Material 3. Motifs-hit matrix.

Motif-hit matrix for each gene was formed from MSigDB collection C3. We removed motifs that have similar sets of target genes.



Supplementary Material 4. Elastic net drug prediction models

Weights (measure of how consistent this protein has been associated with a certain drug (negative values indicate resistance, positive values sensitivity)) from elastic net models for predicting drug response from inferred protein activities.


Supplementary Material 5. Sample source code and sample data sets 


Supplementary Methods

Singular value decomposition of RPPA data


RPPA (phospho) protein expression vectors in the P feature matrix (Pi,1 to Pi,164) are highly correlated. It is therefore desirable to project these feature vectors onto a smaller dimensional space prior to using this data for affinity regression. To achieve this, protein feature vectors of all the proteins are placed in the matrix P and subjected to singular value decomposition (SVD),

P = UPSPVPT,

where UP and VP are the right and left singular matrices and SP is a diagonal matrix whose elements are the square roots of the eigenvalues of the matrix PPT. Only the top 25 dimensions were used for the feature representation, as they account for 92% of the variance.


Affinity regression


We define affinity regression as the following bilinear regression problem. For a dataset of M tumor samples profiled on microarrays with N genes, we let Y RNxM be the mean-centered log gene expression profiles of tumor samples. Each column of Y corresponds to a microarray experiment. We define each gene’s TF attributes in a matrix D RNxQ, where each row represents a gene, and each column represent the hit vector for a TF, i.e. the bit vector indicating whether there is binding site for the TF in the promoter region of each gene. We define the RPPA attributes of tumor samples in a matrix P RMxS where each row represents a tumor sample and each column represents the (mean-centered) log RPPA protein expression profile for the tumor sample. We set up a bilinear regression problem to learn the weight matrix W RQxS on paired of TF-signaling protein features:

DWPT + = Y. (1)

We can transform the system to an equivalent system of equations by reformulating the matrix products as Kronecker products



DWPT (PD) vec(W), (2)

where  is a Kronecker product, and vec(.) is a vectorizing operator that stacks a matrix and produces a vector, yielding a standard (if large-scale) regression problem.

We make the problem more statistically and computationally tractable through a series of steps: we reduce to a smaller system of equations where the output is the set of pairwise similarities YTY between examples rather than Y itself; we reduce the dimensionality of P by SVD; and we use ridge regression to solve for the interaction matrix in the reduced problem. Full details and a derivation of the reduced optimization problem are provided elsewhere (Pelossof et al. 2014). We fit the ridge regression model using the glmnet R package (Friedman et al. 2010) and evaluate performance with 6-fold cross-validation.

Given the protein expression profile of a test tumor sample (centered relative to the mean of the training set), we can right-multiply the protein expression vector through the trained model to predict the similarity of its expression profile to those of the training tumor samples. To recover a reconstruction of the test gene expression profile from the predicted similarities, we assume that the test expression profile is in the linear span of the training profiles. Then a simple transformation converts the vector of computed similarities into a predicted gene expression variation profiles (Pelossof et al. 2014). Finally, to infer the protein activity in a new sample from the (centered) gene expression profile, we can left-multiply through the model via YTDW through the model to obtain a weighting over protein features.


Subtype specificity analysis using mRNA expression levels and protein levels

For the subtype-specific TFs identified through inferred TF activity and listed in Table 1 of the main text, we computed subtype associations directly from m mRNA expression levels and subtypes using the Mann-Whitney U test. Table S1 shows the nominal p-values for TF mRNA levels as well as inferred TF activity associations demonstrating that in general TF-activity is more statistically significant than TF-mRNA levels. For some TFs, subtype-specificity can be also discovered from mRNA levels, but for some TFs it cannot. For example, although MECOM showed similar expression patterns in all subtypes, it has higher transcription factor activity in HER2/LumA/B subtypes. When we look at MECOM’s target genes, we also observed their expression levels were higher in HER2/LumA/B subtypes compared basal.

Similarly, for the subtype-specific signaling proteins identified through inferred protein activity and listed in Table 2 of the main text, we computed subtype associations directly from protein expression levels and subtypes using the Mann-Whitney U test. Table S2 below shows nominal p-values for signaling protein expression levels as well as inferred protein activity associations. For some signaling proteins, subtype-specificity can be also discovered from protein expression levels (especially when the p-value from inferred protein activity is extremely significant), but for some proteins it cannot.
Inferred TF activity and its inferred protein activity association

For each TF in the RPPA dataset, we calculated Spearman rank correlation between its inferred TF activity and its protein expression variation and its inferred protein activity across all tumors. Results are tabulated in the Table S3. For example, for GATA3, we observed a high correlation between its inferred TF activity both protein expression variation and inferred protein activity (Figure S6), suggesting that for this TF, the total protein measurement is informative of its transcriptional regulatory activity. For other TFs, like FOXO3, the correlation between inferred TF activity and protein expression is weak, and the correlation with inferred protein activity is even weaker (Figure S7); the model is learning that this variable is not informative about transcriptional responses. In this case, the issue may be a relatively poor antibody (as flagged by the TCPA website).


Inferred protein activity and RPPA protein expression associations in breast cancer cell lines

Twenty-nine cell lines are available from the TCPA resource that overlap with the Heiser paper (this is largest overlap within the same batch). The affinity regression model downweights protein/phosphoprotein features that are not useful in predicting transcriptional response; therefore, not all inferred protein activities correlated with the original measured values. However, for proteins/phosphoproteins whose inferred activity and measured values are well correlated in the TCGA data set, we would expect to see a similar correlation in the cell line data. Proteins/phosphoproteins whose Spearman correlations between measured and inferred activities were above .35 on the TCGA tumors were used for further analysis. We used the TCGA-trained affinity regression model to predict protein activities and then computed how well they correlated with the actual measured RPPA values from the TCPA resource. Spearman correlations between measured and inferred activities for the proteins considered were again high across the cell line data. In particular, for proteins/phosphoproteins ERα, GATA3, MSH2, AR, Notch3, ERα-pS118, p70S6K, and YB-1-pS102, Spearman correlations between measured and inferred activities are above .35 on the TCGA tumors and are similarly strong for the cell line data. Results are shown in Figure S9.


Transfer Learning with Drug Response Models

For each pair of drugs, we tested how well the drug response prediction model for the first drug predicts the response of the second drug. We trained drug response prediction models across breast cancer cell lines, using inferred protein activities as input features to predict log-transformed GI50 values with ridge regression. If the correlation was less than 0.3, we concluded that the drug’s model cannot predict the other drug’s response and set it to 0 for clearer visualization. Figure S11 shows the heatmap of correlations between predicted and actual drug response, where the drugs in the columns correspond to prediction models, and shows similar relationships to what was observed through the cluster analysis. For example, the model for erlotinib is able to predict response for docetaxel, carboplatin, and cisplatin as well as other EGFR inhibitors including BIBW2992 and gefitinib. Similarly, the model for rapamycin is able to predict response for lapatinib, temsirolimus, 17-AAG and GSK2119563, consistent with the clustering analysis.


Predicting Drug Response with CCLE

The Cancer Cell Line Encyclopedia (CCLE) includes 26 breast cancer cell lines along with drug response data for 24 compounds; 16 out of the 26 cell lines and 7 out of the 24 drugs overlaps with Neve et al. study. We first investigated whether the drug response data of these common cell lines correlate between the two studies. Since GI50 values are not available in CCLE dataset, we used reported AUC values instead. For all drugs except sorafenib, we found a significant positive correlation between the drug response data.
We then asked whether our drug response prediction models, trained on inferred protein activities, generalized to the CCLE breast cancer cell line data. Results are summarized in Table S6. We found that for the 6 drugs with consistent drug response between Neve et al. and CCLE, 5 of the drug prediction models correlated positively with measured drug response on CCLE; 4 of these drugs had fairly good correlations on the CCLE data set. For one drug (topotecan), the prediction model performed poorly despite a good drug response correlation on the common cell lines; we believe the reason may be that the variability of drug responses across the CCLE set is much greater than in the Neve set. Therefore, despite the general problem of inconsistency between large-scale drug response data sets, we found that some of our drug prediciton results did indeed generalize.
Survival Analysis with the NKI and TRANSBIG cohorts

We analyzed breast cancer data from two additional publicly available studies, NKI (n=337)(van de Vijver et al. 2002) and TransBig (n=198) (Desmedt et al. 2007), for which gene expression profiles and overall survival (OS) data were available. We obtained normalized microarray-based gene-expression data from public repositories and used the R package genefu to classify breast cancer samples into the luminal A, luminal B, HER2, normal-like and basal-like molecular subtypes using Pam50 classifiers. For both the TransBig and NKI cohorts, we used (i) the predicted protein activity profiles and (ii) the gene expression profiles corresponding to RPPA proteins to calculate each patient’s risk, and patients were ranked in descending order. (The risk is computed based on hazard functions trained on the METABRIC discovery set; for univariate models, this is just the ranking based on inferred protein/gene expression values.) We designated the top 20% of the patients as the high-risk group and the bottom 40% as low-risk group. The log-rank test was used to compare two Kaplan-Meier survival curves with the null hypothesis that there is no survival difference between the populations. Consistent with results for the METABRIC validation set, Kaplan–Meier survival analysis confirmed that the identified multivariate protein signatures and most of the univariate signatures showed an association with survival in the TransBig and NKI cohorts, while the gene expression signatures for the most part did not validate (Figures S15, Figure S16).


Inferred activity of a specific transcription factor profile associations with functional activation of that pathway

We investigated correlations of copy number and mutational status with inferred protein activity in specific pathways. For example, in non-proliferating cells, RB remains hypo-phosphorylated and forms RB-E2F repressor complexes to inhibit the expression of genes that promote S phase entry. Hypo-phosphorylated RB also directly binds to and inhibits the activity of E2F activators (E2F1-E2F3). RB is inactivated in many human cancers, which occurs through direct mutation or copy number aberrations of RB1 or through a disruption in the regulatory components of the RB-E2F pathway. In our analysis, we observed that E2F1 inferred transcription factor activity was higher when RB1 was deleted (t-test Adj p-value < 10-4) compared to wild-type (Figure S17A). In the TCGA, RB1 is not frequently mutated.



Supplementary References
Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, Wilson CJ, Lehar J, Kryukov GV, Sonkin D et al. 2012. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483(7391): 603-607.

Cancer Genome Atlas Network. 2012. Comprehensive molecular portraits of human breast tumours. Nature 490(7418): 61-70.

Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, Speed D, Lynch AG, Samarajiwa S, Yuan Y et al. 2012. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486(7403): 346-352.

Desmedt C, Piette F, Loi S, Wang Y, Lallemand F, Haibe-Kains B, Viale G, Delorenzi M, Zhang Y, d'Assignies MS et al. 2007. Strong time dependence of the 76-gene prognostic signature for node-negative breast cancer patients in the TRANSBIG multicenter independent validation series. Clin Cancer Res 13(11): 3207-3214.

Friedman J, Hastie T, Tibshirani R. 2010. Regularization Paths for Generalized Linear Models via Coordinate Descent. J Stat Softw 33(1): 1-22.

Heiser LM, Sadanandam A, Kuo WL, Benz SC, Goldstein TC, Ng S, Gibb WJ, Wang NJ, Ziyad S, Tong F et al. 2012. Subtype and pathway specific responses to anticancer compounds in breast cancer. Proc Natl Acad Sci U S A 109(8): 2724-2729.

Li J, Lu Y, Akbani R, Ju Z, Roebuck PL, Liu W, Yang JY, Broom BM, Verhaak RG, Kane DW et al. 2013. TCPA: a resource for cancer functional proteomics data. Nat Methods 10(11): 1046-1047.

Liberzon A, Subramanian A, Pinchback R, Thorvaldsdottir H, Tamayo P, Mesirov JP. 2011. Molecular signatures database (MSigDB) 3.0. Bioinformatics 27(12): 1739-1740.

Neve RM, Chin K, Fridlyand J, Yeh J, Baehner FL, Fevr T, Clark L, Bayani N, Coppe JP, Tong F et al. 2006. A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes. Cancer Cell 10(6): 515-527.

Pelossof R, Singh I, Yang J, Weirauch M, Hughes TR, Leslie C. 2014. Learning the recognition code for transcription factor and RNA-binding protein families from high-throughput data.



van de Vijver MJ, He YD, van't Veer LJ, Dai H, Hart AA, Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton MJ et al. 2002. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347(25): 1999-2009.





1   2   3   4   5


The database is protected by copyright ©dentisty.org 2016
send message

    Main page