Skip to content

Commit

Permalink
update ccle notebook
Browse files Browse the repository at this point in the history
  • Loading branch information
@marouenbg
marouenbg committed Dec 8, 2022
1 parent a4501c1 commit 17ea523
Showing 1 changed file with 117 additions and 13 deletions.
130 changes: 117 additions & 13 deletions netbooks/netZooPy/ccle_analysis.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@
"metadata": {},
"source": [
"# Introduction\n",
"The Cancer Cell Line Encyclopedia (CCLE) has collected various omic data for more than a thousand cancer cell lines, representative of many lineages and tissue type. In this analysis, we will first use DRAGON to find associations between multiomic data types, and second, we will use PANDA-LIONESS-MONSTER to model a transition from primary to metastatic melanoma and identify drivers of this transition.<sup>1</sup>\n",
"The Cancer Cell Line Encyclopedia (CCLE) has collected various omic data for more than a thousand cancer cell lines, representative of many lineages and tissue type. In this analysis, we will first use DRAGON<sup>2</sup> to find associations between multiomic data types, and second, we will use PANDA-LIONESS-MONSTER to model a transition from primary to metastatic melanoma and identify drivers of this transition.<sup>1</sup>\n",
"# Import packages\n",
"First, we start by loading the packages required for the analysis."
]
Expand Down Expand Up @@ -314,7 +314,7 @@
"metadata": {},
"source": [
"# 1. DRAGON multiomic CCLE network\n",
"First, we load the metadata."
"First, we load the metadata that ahs information about cell lines and various omics used."
]
},
{
Expand All @@ -338,7 +338,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"We compute correlations between miRNA levels and gene dependency."
"In the first part, we compute correlations between miRNA levels and gene dependency. Our hypothesis is that strong miRNA repression induces the same effects as gene CRISPR KO."
]
},
{
Expand Down Expand Up @@ -410,7 +410,7 @@
"metadata": {},
"source": [
"## 1.2. Correlations between drug cell viability and gene dependency\n",
"We compute DRAGON partial correlations between drug cell viability and gene dependency."
"Now, We compute DRAGON partial correlations between drug cell viability and gene dependency. Our hypothesis is that drugs inhibits their protein targets and therefore induces similar effects to CRISPR gene KO."
]
},
{
Expand Down Expand Up @@ -448,6 +448,13 @@
"dep_drugs_edges = estimatedepdrug(imputationMissing)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"This plot represents all correlations between gene KO and Dabrafenib cell viability. Dabrafenib is a multikinase inhibitor indicated for melanoma. "
]
},
{
"cell_type": "code",
"execution_count": null,
Expand All @@ -464,15 +471,15 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"We find that gene dependencies correlated with Dabrafenib are BRAF, MAPK1 and MAPK2, which are in the same pathway targeted by Dabrafenib."
"In particular, we find that gene dependencies correlated with Dabrafenib are BRAF, MAPK1 and MAPK2, which belong to the same kinase siganling pathway targeted by Dabrafenib."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1.3. Correlations between LDH protein levels and metabolite levels\n",
"We compute correlations between LDH proteins levels and metabolite levels."
"We compute correlations between LDH proteins levels and metabolite levels. Here, we would like to infer the direction of glycolsis biochemical reactions to see if fermentation (Warburg effect) is prevalent in CCLE cacner cell lines."
]
},
{
Expand Down Expand Up @@ -502,6 +509,13 @@
" return ppi_met_edges"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Now, we plot the correlations between all metabolites and LDH protein levels with its two isozymes (LDHA/LDHB). LDHA carrues the forward reaction for lactate production and LDHB converts lactate to pyruvate, preferentially."
]
},
{
"cell_type": "code",
"execution_count": null,
Expand All @@ -517,15 +531,15 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"We find that metabolites such as fumarate/maleate, PEP, and g3p has a negative correlation with LDHA levels, indicating production of lactate. We also see that LDHB levels have a positive partial correlation (3.705e-05) with lactate which indicates that LDHB works in the same direction as LDHA and further supporting lactate production in cancer cells (Warburg effect)."
"We find that metabolites such as fumarate/maleate, PEP, and g3p have a negative correlation with LDHA levels, indicating production of lactate. We also see that LDHB levels have a positive partial correlation (3.705e-05) with lactate which indicates that LDHB works in the same direction as LDHA and further supporting lactate production in cancer cells (Warburg effect)."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1.4. Correlations between TF targeting scores and metabolite levels\n",
"We first load TF and gene targeting scores for all CCLE cell lines."
"We first load TF and gene targeting scores for all CCLE cell lines. These scores were computed after running PANDA<sup>3</sup> on all CCLE cell lines gene expression to build an aggregate network, then run LIONESS<sup>4</sup> on the aggregate network to build single-sample networks for each cell. We then compute gene and TF targeting scores for each single-sample network."
]
},
{
Expand All @@ -542,7 +556,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"We compute correlations between TF targeting scores and metabolite levels."
"We compute correlations between TF targeting scores and 2HG metabolite levels. Our hypothesis is that we know that 2HG induces a hypermethylator phenotype and a cascade of epigenetic effects, however, we don't know which TF are affected by hypermethylation of their promoters and the consequent change in their binding and activity."
]
},
{
Expand Down Expand Up @@ -586,30 +600,120 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"We find that 2-HG disrupts binding of TP73, PPARg, and GLI4. These TFs have various roles in cancer; TP73 is a tumor supressor, PPARg mediates several oncogenic signaling processes, and GLI4 is a glioma-inducing oncogene. GLI4 is particularly interesting because glioma is the cancer subtype where 2HG induces a hypermethylator phenotype."
"We find that 2HG disrupts binding of TP73, PPARg, and GLI4. These TFs have various roles in cancer; TP73 is a tumor supressor, PPARg mediates several oncogenic signaling processes, and GLI4 is a glioma-inducing oncogene. GLI4 is particularly interesting because glioma is the cancer subtype where 2HG induces a hypermethylator phenotype."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 2. MONSTER transition analysis in melanoma\n",
"In the second part of our analysis, and to follow up on the pan-cancer results we found earlier, we are now interested in a cancer-specific patterns of regualtion, particularly in melanoma. Here, we are interested in transition from primary to metastatic state, to estimate this transition, we will use MONSTER using a LIONESS network of primary melanoma cell line as an initial state, and a LIONESS network of a metastatic cell line as an end state. Since MONSTER in implemented in netZooR, this analysis will run in R.\n",
"First, we define a function that compute a transition matrix, then defines a null distribution by resampling columns of these matrices and recomputing transition 1000 times."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"precomputed=1"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 2. MONSTER transition analysis in melanoma"
"Since resampling can take a while to finish, we can set the `precomputed` tag to load precomputed results."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
"source": [
"if precomputed==0:\n",
" %%R\n",
" runemt <- function(nnet1,nnet2){\n",
" primary219=read.table(paste0(ppath,nnet1,'.csv'),sep=',',header=TRUE, row.names=1)\n",
" metastasis14=read.table(paste0(ppath,nnet2,'.csv'),sep=',',header=TRUE, row.names=1)\n",
" combinedRegNetworks=as.data.frame(cbind(primary219,metastasis14))\n",
" nGenes=length(metastasis14)\n",
" design=c(rep(0,nGenes),rep(1,nGenes))\n",
" monsterResRegNet <- monster(combinedRegNetworks, design ,motif=NA, nullPerms=1000, numMaxCores=12, mode='regNet')\n",
" monsterResRegNet\n",
" }"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We now run the actual analysis, the cell that represents the initial state is [ACH-000580](https://depmap.org/portal/cell_line/ACH-000580?tab=mutation). The final metastatic state is represnted by cell line [ACH-001569](https://depmap.org/portal/cell_line/ACH-001569?tab=mutation) also called MM415. Both cells were sampled from male donors."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"if precomputed==0:\n",
" %%R\n",
" primarycell='ACH-000580'\n",
" metastasiscell='ACH-001569'\n",
" monsterResRegNet=runemt(primarycell,metastasiscell)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We now sort TFs by their differential involvement scores in the transition to metastasis."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"df=pd.read_csv(ppath+'emtrank.csv',index_col=0)\n",
"df.columns = ['score']\n",
"df = df.sort_values(by='score',ascending=False)\n",
"g= sns.jointplot(x=np.array(range(len(df)))+1, y=df[\"score\"], kind='scatter',marker='.')\n",
"plt.text(25,110583,'RUNX2')\n",
"plt.text(7,146068,'GLI1')\n",
"plt.text(7,134995.634187,'CREB3L1')\n",
"g.ax_marg_x.set_axis_off()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We find that RUNX2, GLI1, and CREB3L1 are among the top 50 TFs. These TFs were identified previously in drug resistance (GLI1/CREB3l!) and most importantly in epithelial to mesenchymal transition (RUNX2)."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# References\n",
"\n",
"1- Guebila, Marouen Ben, et al. \"The Network Zoo: a multilingual package for the inference and analysis of biological networks.\" bioRxiv (2022)."
"1- Guebila, Marouen Ben, et al. \"The Network Zoo: a multilingual package for the inference and analysis of biological networks.\" bioRxiv (2022).\n",
"\n",
"2- Weighill, Deborah, et al. \"DRAGON: determining regulatory associations using graphical models on multi-omic networks.\" arXiv preprint arXiv:2104.01690 (2021).\n",
"\n",
"3- Glass, Kimberly, et al. \"Passing messages between biological networks to refine predicted interactions.\" PloS one 8.5 (2013): e64832.\n",
"\n",
"4- Kuijjer, Marieke Lydia, et al. \"Estimating sample-specific regulatory networks.\" Iscience 14 (2019): 226-240.\n",
"\n",
"5- Schlauch, Daniel, et al. \"Estimating drivers of cell state transitions using gene regulatory network models.\" BMC systems biology 11.1 (2017): 1-10.\n",
"\n",
"6- Cohen‐Solal, Karine A., Howard L. Kaufman, and Ahmed Lasfar. \"Transcription factors as critical players in melanoma invasiveness, drug resistance, and opportunities for therapeutic drug development.\" Pigment cell & melanoma research 31.2 (2018): 241-252."
]
}
],
Expand Down

0 comments on commit 17ea523

Please sign in to comment.