Data Normalisation in ResistoXplorer

Hi. And welcome back to another video. In this video, we will learn how we can normalise resistome data using ResistoXplorer. So by far in the course, we have already seen how to upload and perform data filtration with our resistome count table in ResistoXplorer. So in this video, we are going to directly jump to the data filtration. And it’s going to skip those pages. As you can see, this is the data normalisation page which will appear once you have done data filtration.

In terms of design, the speech looks very similar to data filtration page, as we have on the top end the detailed text information, while in the bottom half we have a box which contain a variety of actual option of normalising the data. So the main aim of normalising is to standardise the data in order to make accurate comparisons between the samples. In ResistoXplorer, we have implemented three different approaches of normalising the data, the data rarefying, scaling, and the data transformation. So when talking about data rarefying, this method is only used when there’s a big disparity in the library size as it may lead to loss of valuable information.

You can always go to the data inspection page and the library site overview in order to check the sample size or library size overview for each sample, though this approach is still commonly used and it performs overall better in some of the downstream analysis. And that’s why it has been still implemented in ResistoXplorer. Apart from this, we have a variety of other scaling and transformation-based methods which are very commonly used in metagenomic data analysis, such as total sum scaling. And then we have another one which is derived from the metagenomic R package called cumulative sum scaling.

Also, we have supported some commonly used variant-stabilise transformation, such as relative log expression and trimmed mean of M-value, which are originally adopted from RNA-seq count data analysis. However, most of these methods do not account for data compositionality. As a result, we have implemented centred log ratio and additive log ratio transformation, which are called alternative for normalising the data and addressing the issue of data compositionality. One important thing to note here is that in ResistoXplorer we cannot perform data scaling and data transformation together as the count data will no longer remain count once we applied one of these methods. So once we choose one of the methods the other will automatically set to do not transform or do not scale.

In this video, I’m just using the default parameter and clicking on Submit button to perform data normalisation.

As you can see, we will have a message board up here in the upper right corner telling whether the data normalisation is OK or if there is any error. If the data normalisation is OK it will provide you details which normalisation have you selected. In addition to this, you can see in the bottom right corner the Proceed button also will be enabled. And we can click on this Proceed button to go to the next step, which is the analysis panel, which contains a variety of downstream analysis modules such for performing composition profiling, clustering analysis, differential analysis, and machine learning or biomarker prediction in order to gain insight from our uploaded assistant count data.

We will explore all these modules one by one in the upcoming lectures and video. For now, try uploading your own data set or any example data set and perform data normalisation steps. Additionally, you can choose a variety of different normalisation steps and assess the effect of this normalisation approach using the visual data exploration and the clustering analysis page, such as coordination analysis and heat map and dendrograms. That’s it for this video. Until then bye-bye, and take care. We’ll see you very soon. Bye.