Found 2 recipes
Does my RNA-Seq data contain noticeable batch effects? Can I remove any batch effects I find?
This recipe provides one method to assess RNA-seq count data for potential batch effects. Identified batch effects are removed using the ComBat algorithm (Johnson, Rabinovic, and Li). An example use case of this recipe is when an investigator wants to correct data and account for confounding variables generated by effects such as using different reagents or processing samples at different times.
This recipe uses count data and sample annotations derived from two separate RNA-seq experiments and relies upon Principal Component Analysis (PCA) as a primary means of evaluating the dataset for batch effects. In particular, this recipe utilizes a number of GenePattern modules to filter and preprocess the RNA-seq count data and to normalize the dataset using the ComBat algorithm. We then use Galaxy to view the influence of batch effects in our pre- and post-normalization datasets.
Why consider batch effects (a.k.a. confounders)? A "batch" or "confounding" variable can be generally defined as an extraneous variable that may correlate or anti-correlate with our variables of interest. Failing to account for confounding variables or batch effects can lead us to draw spurious conclusions from our data. For example:
“Albert is comparing the hardiness of two different E. coli strains, X and Y, by measuring their production of various stress proteins under a mild bactericide. However, Albert's bactericide stock is running low and he ends up using two different stocks of the same bactericide over the course of this experiment. Here, bactericide stock is a confounding or batch variable. Perhaps one bactericide stock is more potent than the other. Albert will need to be careful in how he analyzes his data. He must somehow ensure that differences in stress protein abundances between his X and Y strains are driven by the biological differences of these two strains rather than differences in the bactericide stocks.”
Efforts can be undertaken to minimize batch effects by ensuring proper experimental design, e.g., simplifying protocols to eliminate potential confounders. If batches are unavoidable, it is also important to randomize samples across these batches such that differences between batches can be attributed to batch effects rather than biological differences. But despite these efforts, unforeseen batch effects may still be present in any experiment. Thus, we should be aware of the degree to which these unwanted factors influence our results and be prepared to use various statistical methods to remove these unwanted factors if needed.
How do we identify batch effects? There are a variety of methods for identifying batch effects. Sometimes, potential batches can be identified at the outset of a project based on its experimental design. For example, if samples must be processed in separate groups due to the limiting capacity of a sequencer, then these processing groups are likely to introduce batch effects. If batches can be pre-identified, then the scientist can include identical control samples in each batch. Any variation in measurements from these control samples can then be attributed to and used to quantify a batch effect.
Another common method is to visualize data using a PCA projection. PCA uses a linear transformation to transform a dataset into a space of linearly uncorrelated, orthogonal "principal components". More simply, PCA is a way of visualizing data such that underlying structures are revealed. We encourage users to learn more about PCA (Ma and Dai's 2011 article in Briefings in Bioinformatics is a good starting point). By re-projecting with PCA, we can identify those variables, including batch variables, that are contributing to large variances in the dataset.
What genes are essential to a cell’s survival in a specific environment?
This recipe provides a way to process the results of genome-wide CRISPR-Cas9 knockout screens. In these screens, single guide RNAs (sgRNAs) are designed to bind to and inhibit specific target DNA sequences in genes. Multiple sgRNAs may target the same gene to increase knockout efficiency. In positive screens, essential genes are identified through the sequencing of surviving cells post-selection. The loss of these ‘winning’ genes create cells that are resistant to the selective pressure. In negative screens, essential genes are identified by measuring which genes are lower in abundance post selection. These screens require a non-selected control, which is used to find which genes are essential to survival under the given selective pressures (Miles et al., 2016). Since a large number of sgRNAs can be introduced in a single screen, many genes can be tested for a selection criteria. However, there are many factors to consider in processing of sequenced reads; often multiple sgRNAs in a library target the same gene but with different specificities and efficiencies, and read count distributions vary depending on library and study designs. Additionally, positive selection screens often result in relatively few sgRNAs that dominate the total sequenced reads. The MAGeCK (Li et al., 2014) method was specifically developed for CRISPR screen analyses with these conditions in mind.
How can we find the molecular mechanism responsible for resistance?
By looking at how the hits in the screen aggregate on an interaction network, we can get an idea of the mechanisms that are essential for the organism to survive an environmental challenge. The network neighborhood that contains a high concentration of essential genes is strongly implicated as the molecular mechanism by which an organism handles the challenge.
We can find the network neighborhood that is enriched for the screen hits through an algorithm called network propagation (Carlin et al., in press) that is implemented as a feature of the popular network analysis program Cytoscape. This algorithm will find the closely clustered hits and their network neighbors to build a network diagram of the resistance mechanism. We can then use GeneMANIA plugin to find enriched terms that easily summarize the biological terms that are enriched in the diagram.
What is Model-based Analysis of Genome-wide CRIPSR/Cas9 Knockout (MAGeCK)?
Model-based Analysis of Genome-wide CRIPSR/Cas9 Knockout (MAGeCK) is an algorithm for identifying both positively and negatively selected sgRNAs and genes from genome-scale CRIPSR/Cas9 knockout screens. The MAGeCK method can be summarized by the following steps:
1. sgRNA read counts are median-ratio normalized.
2. Mean-variance modeling is then used to model each replicate. The statistical significance of each sgRNA is calculated using the learned mean-variance model.
3. Essential genes are determined by looking for genes with consistently highly significant sgRNAs using robust rank aggregation.
Use Case: MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens (Li et al. 2014).