 Hello, my name is Mihal Bordukiewicz and I will talk about air as an environment for the functional analysis of proteins. So with the outline, we are actually skipping the whole thing because of the functional analysis of proteins we want to discuss in details, what we'll focus on two phases of our model, development and deployment. First and foremost, I am a scientist, so I want a model which will be a functional model that will be seen as a supplement to the publication when I'm presenting its pros and cons. So I'm trying to develop models in a very reproducible manner, but also I want these models to be easy to share and to deploy. So there's few words about functional analysis of proteins. In this very case, we are trying to use biological data, in my case, mostly primary sequences of proteins, so the vectors of amino acids, to create a model which is predicting, which is predicting in a supervised manner some labors of these sequences. I know this sequence before the training. I know this labors before the training. So let's go. So how to develop such model? Firstly, we need to process our sequence and there are multitude of packages that could easily process your biological sequences. Among the current-based packages, I want to highlight APE, BIOSEC and SAKINAR, especially BIOSEC. They are allowing easy processing of biological sequences and BIOSEC is in a very tidy way. And also our package TIDSQ, currently on Github, is able to do this prediction in a very fast, very efficient manner because of the lot of C++ under the hood. So later we are developing the models and here we cannot really do much besides using one of these free, excellent packages. We sometimes decide to opt out from any of them when we have a model which is a bit more complicated, but still when we are looking for a potential model, they are streamlining our work. And the last issue is our reproducibility, the reproducibility of the model. So please use targets or use the former version of Drake and Renf to ensure that your code can be run by others. And one of the examples could be the code of one of my software packages, Amprogram. It will show here and there in this publication further. So we have our great model and it's time to deploy it. So because in the functional analysis of proteins, we are often aiming at biologists which are less affluent, I like the most making these models available as shiny web applications. So I am doing the shiny web app. The important part is that either the biologist can paste sequences here in this field or he can submit the FASTA file from its hard drive. So there are two ways of submitting the data to make it a bit more convenient. And of course, example is the shiny app of the Engram. And one last part is the deployment of the R package because if you are trying to make a package with your model, you are of course creating a predict function that will work with your train model and so on. And it all seems easy but sometimes models are too large for Cran because Cran accepts only packages up to the size of five megabytes. In this case, you can make an external dependency, external GitHub package with model and so on Cran we are hosting the Engram package, but the Engram package is depending on the Engram model which can be downloaded from the web. Please check our model, Engram model and Engram to see how it's done. And this is all the time that I have so I want to thank my group and if you need any help with that, please reach me using my email. Bye bye.