Supervisors:
- Nicolas Palacio ([email protected])
- Aurelien Dugourd ([email protected])
Tutor:
- Alvaro Mendoza ([email protected])
Cancer is a generic term for a heterogeneous group of diseases that arise in different parts of the body. Cancer stems from genetic alterations that transform normal cells into malignant cells. These malignant cells are characterized by an uncontrolled growth and resistance to apoptosis (programmed cell death). This malignant behavior is mainly achieved through genetic alterations that affect the activity and wiring of many molecular mechanisms like signaling and metabolic pathways. Therefore cancer cells hijack many cellular processes enabling them to proliferate uncontrollably, avoid apoptosis and/or immune response, migrate to other tissues (metastasis) or even obtain drug resistance. Moreover, large-scale studies from international cancer consortia have described that patients' tumors acquire this malignancy through diverse individual molecular alterations. In other words, individual cancers are different each other.
Lung cancer is one of the most common cancer types in the world and with high mortality rates. Many studies have been already published showing potential targets and already approved drugs are being used to treat this disease. Nevertheless, due to the inherent heterogeneity of cancer and patients, finding an easy way to characterize potential response or resistance to these treatments is key to counteract the malignancy. This is especially true in patients with very aggressive tumors, where time is key to maximize a positive outcome. To tackle this problems, generating and analyzing comprehensive biological data has shown to be the best approach.
You will be analyzing transcriptomic data from lung cancer patients (as well as normal lung samples for control) in order to find potential markers to stratify tumor types, stages and potential drug targets based on this expression data.
NOTE: The following exposed points are a suggestion, you are free to perform your analysis using other tools or explore different aspects of the data set. Just keep in mind what MUST be part of the project and that whatever method you are applying has a clear purpose and justification (i.e. do not apply a tool or method just because you found it, but make sure it makes sense in your context and serves a clear purpose).
First, familiarize and explore the data by reviewing literature and performing some basic descriptive statistics.
- Look up in the literature to find more about the data you are analyzing, possible analyses that can be performed and/or what results you can expect.
- Explore gene expression profiles in cancer and healthy samples (e.g. density plots, violin or box plots, profile/correlation heatmap...).
- Observe what type of data, which scale and ranges you are working with.
Are the values raw gene counts, intensities? Are they in logarithmic scale?
- Assess whether if data needs to be normalized or if there is any experimental bias (i.e. batch effect) across samples that should be addressed.
Do they look already normalized? Could you identify number of batches from the data if any?
Explore the space of gene expression using dimensionality reduction methods such as PCA, k-means, t-SNE or UMAP are among the most common.
- Relate the main latent factors to the sample groups (i.e. cancer types).
Is there any co-variate that correlates with the main aspects of transcriptional heterogeneity in the data set?
For this part of the analysis, you can use the metadata of the samples from the original study. This table is provided as a comma-separated file (csv) file along with the expression data of the project (cf. Description of the data set section).
Once you are familiar with the data, you can then address more specific questions. First, choose a subtype of lung cancer of your interest to study in further detail. Each group should choose a different subtype out of the following:
- Adenocarcinoma
- Squamous cell tumors
- Large cell neuroendocrine
- Basaloid tumors
- Carcinoid tumors
- Small cell carcinoma
Once chosen, a good place to start your specific analysis is by performing a differential expression analysis. This allows you to extract which genes are significantly up- or down-regulated in a given comparison, for instance, cancer subtype vs. healthy.
- Perform a differential expression analysis and assess the significance using t-test (one test per each gene). Samples of the same group can be considered replicates (e.g. all healthy samples, all samples of a cancer type...).
Note that differential expression values are expressed in terms of fold change (FC). This is the ratio between the comparing conditions (e.g. value of gene A in condition X divided by the value of gene A in condition Y). Also note that when working in logarithmic scale this is done by subtraction of the log-values:
log(FC) = log(X/Y) = log(X) - log(Y)
Once the significant changes are obtained, you may extract molecular or functional signatures to identify relevant processes (e.g. signaling pathways) from the data. One common approach is the use of enrichment methods, like Gene Set Enrichment Analysis (GSEA).
- Extract relevant signatures from the results of the differential expression of your contrast(s) of interest.
You can enforce or assess your results by contrasting them with the literature.
With your previous results you may have identified some genes or gene sets of interest in your comparison. You can then try to fit a linear model of these changes in increasing tumor stages or other covariates you find interesting like metastatic and non-metastatic tumors.
Data source in GEO: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30219
The expression data set and the metadata of the samples is provided in the data folder of the repository. Please use this for your analyses as the probe identifiers have been already translated to gene names and the samples are named according to the cancer subclass instead of GEO identifiers.
- expression_dataframe.csv: Here you will find the expression data of the different samples and transcripts. Rows are the gene names and columns are the different samples according to the lung cancer subtype.
- sample_infos.csv: The metadata of the samples. You can find the name of the sample as provided in the expression_dataframe.csv file, the gender, age and status of the patient, the TNM stage of the different tumors (see Literature for more information), if the tumor has relapsed and the follow-up time of the patients (in months), smoking habits and if the patient has received chemo-/radiotherapy.
Data set comprises 293 lung tumor samples and 14 non-tumoral lung samples. Tumor samples come from different subtypes of lung cancer and stages.
| Subclass | Abbreviation | Samples |
|---|---|---|
| Non-tumoral lung | NTL | 14 |
| Adenocarcinoma | ADC | 85 |
| Squamous cell tumours | SQC | 61 |
| Large cell neuroendocrine | LCNE | 56 |
| Basaloid tumours | BAS | 39 |
| Carcinoid tumours | CARCI | 24 |
| Small cell carcinoma | SCC | 21 |
| Other histology | Other | 7 |
| Total | 307 |
- Original article here
- Supplementary data here. You may find specially interesting for this project the supplementary table 7.
You first task will be to define a project proposal, which should include:
- Summary of literature on this data set
- Questions you want to address
- Approximate timetable
You will present this project proposal together with a literature review on the subject 3 week after the beginning of the semester (10 minute presentation + 5 minutes discussion).
You project MUST contain the following elements:
- Descriptive statistics about the data sets
- Graphical representations
- Dimension reduction analysis (PCA, clustering or k-means)
- Statistical tests (t-test, proportion tests etc)
- Linear regression analysis, either uni- or multivariate