in prior videos of the bioinformatics from scratch series you have learned how to compile your very own bio activity data sets directly from the chambo database how to perform exploratory data analysis on the computed lipinski descriptors you have also learned how to build random forest model as well as building several machine learning models for comparing the model performance using the lazy predict library and so in this video we will be taking a look at how we can take that machine learning model of the bioactivity data set and convert it into a web application that you could deploy on the cloud that will allow users to be able to make predictions on your machine learning model for the target protein of your interest and so without further ado we're starting right now okay so the first thing that you want to do is go to the bioactivity prediction app folder and so this folder will be provided in the github link in the video description and so before we start let me show you how the app looks like so i'm going to activate my condy environment and for you please make sure to activate your own content environment as well so on my computer i'm using the data professor environment so i'm going to activate it by typing in conda activate data professor and i'm going to go to the desktop because that is where the streamlet folder resides and then we're going to go to the bioactivity folder let's have a look at the contents so the app. py will be the application and so we're going to type in streamlit run app. py in order to launch this bioactivity prediction app okay and so this is the bioactivity prediction app that i'm going to be showing you today how you could build one and so let's have a look at the example input file so this is the example input file so in order to proceed with using this app we're going to have to upload the file drag and drop right here or browse files and select the input file and so while waiting for a input file to be uploaded you can see here that the blue box will be giving us a waiting message so it's saying upload input data in the sidebar to start so essentially the input file contains the smiles notation and the chambo id and so the tempo id you can think of it as kind of like the name of the molecule here and particularly the tempo id is a unique identification number of this particular molecule that chambo database has assigned to it and the smile notation here is a one-dimensional representation of this particular chemical structure and so this my instantation will be used by the paddle descriptor software that we're going to be using here today in the app in order to generate molecular fingerprint which describe the unique chemical features of the molecule and then such molecular fingerprints will then be used by the machine learning model to make a prediction okay and so the prediction will be the pic50 values that you see here and the psc 50 value is the bioactivity against the target protein of interest and so in this application the target protein is acetylcholinesterase and this target protein is a target for the alzheimer's disease okay and so this app is built in python using the trim-lit library and the molecular fingerprints are calculated using the paddle descriptor so back in 2016 we have published a paper describing the developments of a q-star model for predicting the bioactivity of the acetylcholinesterase and so if you're interested in this article please feel free to read it and so i'm going to provide you the link in the video description as well okay so let's drag and drop the input file so example esthetically in series i'm going to drag a drop here and then in order to initiate the prediction i'm going to press on the predict button and as you see here the input file is giving you this data frame and then it's calculating the descriptor and the calculated descriptor is provided here in this particular data frame so you're going to see here that there are a total of five input molecules and there are 882 columns and you're going to see here that the first column is the tempo id so in reality you're going to have a total of 881 molecular fingerprints and the molecular fingerprints that we're using today is the pubchem fingerprint and because we have previously built a machine learning model which i will be showing you using this file the jupyter notebook file we had reduced the number of descriptors from 881 to 217 no actually 218 because we have already deleted the first column the name of the the symbol id column and so we have reduced from 881 columns to 218 columns okay and so in the code we're going to be selecting the same 218 columns that you see here which corresponds to the descriptor subsets from the initially full set of eight eight one okay so we're going to use the 218 as the x variables in order to predict the psa 50 and finally we have the prediction output and the last data frame here and we have the corresponding tempo id and then we could also download the prediction by pressing on this link and then the prediction is provided here in the csv file okay so the data is provided here all right and so let's get started shall we okay so we have to first build our prediction model using the jupyter notebook and then we're going to save the model as a pickle file right here okay so let me show you in which will take just a moment so let me open up a new terminal and then i'm going to activate jupyter typing in jupyter notebook okay so i have to first activate condy environment kind of activate data professor so it's the same environment and then jupyter notebook all right there you go and then i'm going to open up the jupyter notebook all right and here we go so actually this was adapted from one of the prior tutorials in this bioinformatic from scratch series and essentially we're going to just download the calculated fingerprints from the github of data professor using this url link and so we're importing pandas as pd and then we're downloading and reading it in using pandas and the resulting data frame looks like this and so you're gonna see here that we have all of this so one column the last column is pic50 and we have 881 columns for the pubchem fingerprint and then the next cell here is we're going to be dropping the last column or the pic50 column in order to assign it to the x variable and then we're going to just select the last column denoted here by -1 and assigning it to the y variable and so now that we have the x and y separated we're going to next remove the low variance feature from the x variable so initially we have 881 and so applying a threshold of 0.
1 this resulted in 218 columns and then we're going to be saving it into a descriptor list. csv file so let me show you that descriptor lists the csv file okay and then you're going to see here that the first row will contain the names of the fingerprints that are retained in other words the name of the descriptors of the 218 columns here we here you can see that pop can fingerprint 0 1 2 has been removed and we have fingerprint 3 and fingerprints 4 until 11 has been removed fingerprint 14 has been removed fingerprint 17 has also been removed so more than 600 fingerprints have been deleted from the x variable and so the removal of excessive redundant features will allow us to build the model much quicker okay and so in just a few moments i will be telling you how we're going to be making use of this descriptor list in order to select the subsets from the computed descriptors that we obtained from the input query right here let me show you that we get from the input query right here so out of this small citation we generated 881 columns and then we're going to be selecting a subset of 218 from the initially 881 by using this particular list of descriptors okay and let's go back to the jupyter notebook all right let's save it and then we're going to be building the model random forest model we're setting here the random states to be 42 the number of estimators to be 500 and we're using the random force regressor and we fit the model here in order to train it and then we're going to be calculating the score which is the r2 score and then we're assigning it to the r2 variable and then finally we're going to be applying the trained model to make a prediction on the x variable which is the training sets also and then we're assigning it to the wide red variable okay so here we see that the r squared value is 0. 86 and then let's print out the performance mean squared error of 0.
34 and let's make the scatter plot of the actual and predicted values okay so we get this plot here and then finally we're going to be saving the model by dumping it using the pickle function pico dot dump and then as input argument we're going to have model and then we're going to save it as essential calling series model dot pkl and there you go we have already saved the model okay so i'm going to go ahead and close this stupid notebook and let's help over back and let's take a look at the app. py file okay so let's have a brief look you're going to see here that the app. py is less than 90 lines of code and about 87 to be exact and you're going to see that there are some white spaces so even if we delete all the white space it might be even less maybe 80 lines of code okay so the first seven lines of code will be importing the necessary libraries and so we're making use of streamlit as the web framework and we're using pandas in order to display the data frame and the image function from the pil library is used to display this illustration and the descriptor calculation will be made possible by using the subprocess library so that will allow us to compute the title descriptor via the use of java and we're using the os library in order to perform file handling so here you're going to see that we're using the os dot remove in order to remove the molecule.
smi file so i'm going to explain to you that in just a moment base64 will be used for encoding decoding of the file when we will make the file available for download the prediction results and the pickle library will be used for loading up the pickled file of the model okay and so you're going to be seeing here that we're making three custom functions so lines 10 through 15 the first custom function will be our molecular descriptor calculator so we're defining a function called desk calc and then the statement underneath it will be the batch command and so this batch command is what we're normally using when we type into the command line okay and so this option here will allow us to run the code in the command line without launching a gui version of paddle descriptor and so without this option here it will launch a gui version but since we don't want that to happen we're going to use this option okay and so we're using the jar file to make the calculation of the fingerprints and then you're going to see here that we have additional options such as removing salt standardizing the nitro group of the molecule and then we're using the fingerprint to be the pubchem fingerprint using the xml file here and then finally we're generating the molecular descriptor file by saving it to the descriptor's underscore output. csv file and so this batch command will be serving as input right here in the subprocess. p open function okay and then finally after the descriptor has been calculated we're removing the molecule.
smi file and so the molecule. smi file will be generated in another function so i will be discussing that in just a moment and the second custom function that we're generating here is file download so after making the prediction we're going to be encoding decoding the results and then the output will be available as a file for downloading using this link and the third function that we're creating is called build model so it will be accepting the input argument which is the input data and then it will be loading up the pickle file which is the built model into a load model variable and then the model which we have loaded will be used for making a prediction on the input data which is specified here and after a prediction has been made we're going to be assigning it to the prediction variable then we're going to be printing out the header called prediction outputs which is right here and underneath it we're going to create a variable called prediction output and we're going to be creating a pd dot series so essentially it is a column using pandas and so the first column is prediction and then we're naming it pic50 which is here and then we're going to create another variable called molecule name and the column that we're creating is the chamber id or the molecule name which is right here the first column and then we're going to be combining these two columns given by the individual variables called prediction outputs and molecule name so we're using the pd. concat function and then in bracket we're using molecule name which is the first column prediction output which is the second column and then we're using an axis equals to one in order to tell it to combine the two variables or the two columns in a side-by-side manner okay so axis one will allow us to have the two columns side by side otherwise it will be stacked underneath it so psv50 column will be stacked underneath the molecule name if the axis was to be zero okay and finally we're writing out the data frame which is here and then we're allowing it to generate the download link which is right here and we're making use of the file download function described earlier on here okay and then aligns number 38 we're generating this or displaying this image of the web app okay and lines number 43 until 51 or 52 is the header here the bioactivity prediction app title and then the description of the app and then the credits of the app and this is written in markdown language all right and so let's have a look further lines 55 until 59 will be displaying the sidebar right here so 55 will be displaying the header number one upload your csv data and then we're creating a variable called uploaded file and here we're using the st.
sidebar dot file loader file uploader and then as input argument or displaying the text upload your input file which is also right here and then the type of the file will be the txt file so right here and then we're creating a link using markdown language to the example file provided here to the example essential coding series so it's going to be the exact same file that we have selected as input okay so that's the sidebar function that you see here all right and so let's have a look for so here you can see that from line 61 until 87 we have the if and else condition so if we click on the predict button which is right here using the st. sidebar dot button function with input argument of predict so if we click on it it will make the descriptor calculation and apply the machine learning model to make a prediction and finally displaying the results of the prediction right here and allow the user to download the predictions however if we didn't click anything whereby we loaded up the web page from the beginning as i will show you right now you will see a blue box displaying the message of upload input data in the sidebar to start okay so two conditions if the predict button is clicked it will make a prediction otherwise it will just display the text here that it is waiting for you to upload the input data okay so let's have a look under the if condition so upon clicking on the predict button as you have guessed it will load the data that you had just drag and dropped and then it will be saving it as a molecule. smi file and this very same file here molecule.
smi will be used by the desk calculation function that we have discussed earlier on particularly the molecule. smi file will be used by the paddle descriptor software for the molecular descriptor calculation and after the descriptors have been calculated we will assign it as the x variable it's right here okay so i'm going to tell you in just a moment lines number 65 will be printing out the header right here so let me make a prediction first so that we can see let's drag and drop the input file [Music] press on the predict button it's right here original input data line number 65. line number 66 will be printing out the data frame of the input file so you're going to see here two columns the smile citation which represent the chemical structure information and the tempo id column lines number 68 will be displaying a spinner so upon loading up this results here by pressing on the predict button you saw earlier on that they had a yellow message box saying calculating descriptor and so underneath we have the desk calculation function and after it is calculated it will be displaying the following content the calculated molecular descriptor which follows here on lines number 72 right here calculated molecular descriptor and then it will be reading in the calculated descriptor from the descriptor output.
