This is the 6th and (probably?) final post in my series Traversing TCGA.
So far, I have downloaded, extracted cleaned and analyzed data from The Cancer Genome Atlas for acute myeloid leukemia (AML) patients. I have used python to perform linear regressions of patient survival against various clinical data.
Now I have made it available as an interactive web application.
This application allows users to select which features to include in the model, then Submit. A predictive model with is generated with linear reagression and the predicted survival is plotted vs. actual. If the prediction is good, there is a positive correlation. If not, there is little or no correlation.
The web tool uses a PostgreSQL database to store data. In this case, the size of the data is small, and it likely won’t be updated regularly, so a relational database (RDB) like Postgres is not really necessary. The data could simply be stored as a .csv or .xml on the server instead. However, I decided to use Postgres to teach myself how it works and can interact with a web application.
The web page itself uses checkboxes to allow users to select which features to include in the model. Each checkbox has an attribute (‘value’) which uniquely corresponds to the name of a column in the database. Upon clicking submit, the user’s browser sends a list of all these values to the server. The server queries the database for the relevant feature columns and the target variable (survival). Then the server performs a linear regression fit and sends the predicted and actual survivals to the browser.
The communication between browser and server occurs via AJAX, handled by a jQuery.getJSON() request, and python Flask, listening on the server. The Flask app activates Python code to run–scikit learn for statistics and psycopg2 for querying the Postgres database–and generate the linear regression and send the data back to the browser, serialized via JSON. The user’s browser recieves the data and plots it using Plotly.
Please give the app a try. Comments/questions are very welcome.
Beyond Data Science 