sandrakla / adri

Shiny App: Calculation of Age-dependent Reference Intervals (AdRI)

Hello 👋

My name is Sandra and I was a student at the TU Braunschweig (Master of Biology, finished Sept 2020) with a focus on Bioinformatics. A broad and interesting field that brings new challenges every day! 🌱🔬

I am particularly interested in age-dependent reference intervals. I have developed several Shiny Apps on this topic:

• AdRI/AdRI_GAMLSS
• AdRI_rpart
• Zlog_AdRI
• AdRI_Generator
• reflimR_Shiny

I have been working on a Shiny App for processing data from the FreeSurfer program for analysing MRI images of the brain (see ShinySurfer). At work I use databases and SQL to create various GUIs in C# (.NET Framework) and data analysis 💻. In my free time I look at some HTML, CSS and Javascript to create websites and learn Python 🌎.

Over time, I have gained a lot of experience in data science, software development and machine learning. Here is a list of my projects and publications .

📫 How to reach me: You can leave me a message via GitHub Discussion!

"Essentially, all models are wrong, but some are useful" - George E.P. Box

adri's Issues

Check format of uploaded datasets

Combine Regressions with the age groups in the Window Method (LIS, Decision Tree + Sliding Window)

Download data with year not only in days

To change the age in the tables (e.g. 3.75 years) to Year/Months/Days (e.g. 3 Years 9 Months 0 Days) use the following functions in Excel:

=INT(A1) & " Years " & INT((A1-INT(A1))*12) & " Months " & ROUND((((A1-INT(A1))*12)-INT((A1-INT(A1))*12))*30;0)& " Days"
=GANZZAHL(A1) & " Years " & GANZZAHL((A1-GANZZAHL(A1))*12) & " Months " & RUNDEN((((A1-GANZZAHL(A1))*12)-GANZZAHL((A1-GANZZAHL(A1))*12))*30;0)& " Days" (GERMANY)

Verification that the zlog value is calculated correctly

Optimize the axis

• x-scales of the graphs are displayed by default in the unit years or logarithmic

Simplification of the tables

to

Future: Include possibility to split the dataset by origin (STATION) and gender (SEX)

Add Confidence Intervals

Coloring of the Decision Tree Plots

Optimize Sliding Window Method

• Use age groups and not index to iterate through the data
• Add Metrics to the Sliding Method

Dynamically SliderInput

See https://groups.google.com/g/shiny-discuss/c/8wboYTfo6ek

Optimize Regressions

• Decision Tree for optimal age group portioning
• Piecewise Regression (see Virtanen et al. (2020))
• Creation of a connection to Zlog_AdRI

tabPanel("Analysis",

  #          
  #   p("Linear Regression:"), 
  #   plotOutput("regression_stats_linear", height = "500px"),
  #   p("Polynomials (Degree 10):"),
  #   plotOutput("regression_stats_poly10", height = "500px"),
  #   p("Polynomials (Degree 3):"), 
  #   plotOutput("regression_stats_poly3", height = "500px"),
  #   p("Polynomials (Degree 4):"), 
  #   plotOutput("regression_stats_poly4", height = "500px")
  # )

Get automatically the unit from the data

Hyperparameter optimization

• Decision Tree in the Window-Methods
• gamlss ( sliderInput("epochs", "Number of Epochs:", 5 , 250, 30),) -> find.hyper()
• Regressions

Optimize Density Plot

• like plotSimplegamlss()
• #tabPanel("2D Density Plot", plotlyOutput("hexbinplotly", height="600px")),

Hide/enable tabs if the models have not yet been created

Complete revision of the comparison of window methods

Comparison

The Regular Window-Method and Window-Method coupled Decision Tree can be compared with the metrics:

R-Squared (R2)
Mean Absolute Error (MAE)
Mean squared error (MSE)
Root mean squared error (RMSE)

zlog + Window-Method

Future: adding more methods, e.g. quant regression?