If you have 3 responses, height of plant, biomass, sugar content and multiple predictor variables, pH, Mineral content, humus, region/plot, etc, why not try a multiple linear regression or a multivariate regression? 

Are the soil balls taken from the same area/plot? If so, you can add in a repeated measures component to the analysis.  

Thank you for the answer!

Four soil balls with the plants were collected from diffferent regions. 

Can I do that if the predictor variables (pH, humus etc.) are exact numbers?

Hello Dávid Nagy,

I believe you need first to determine if all of your dependent variables indeed have a correlation with the independent variables for building a model.

For that I prefer the Maximal Information Coefficient (MIC), where all information are enable in the link and publication below. It will define how the variables are related and provide a selection variable criteria for your model.

About the model itself, I believe working with plants (the real world!) is not indicate a linear method, e.g., multiple and multivariate linear regression. Maybe you can use a Multilayer Perceptron which deal with nonlinear relations very well.

In the links below you can find all necessary information about R package that implements Multilayer Perceptron. It is not hard use them in R. There is a page in the exploredata website for downloading all necessary script for executing MIC into R.

http://www.exploredata.net/

https://cran.r-project.org/web/packages/RSNNS/index.html

Article Detecting Novel Associations in Large Data Sets

Hello Lucas Venezian Povoa!

Thank you for your answer! I will check these articles and try to use in my case! 

Thanks again!

Best!

David