------------------------------------------------------------

Random Forest Model Objects for Pulmonary Toxicity Risk Ass-
essment

Jeremy M. Gernand

15 April 2013

------------------------------------------------------------

http://nanohub.org/resources/17539

------------------------------------------------------------

This download contains MATLAB treebagger objects, random 
forest models, based on a meta-analysis of published 
pulmonary nanoparticle toxicity experiments. 

There are 5 individual model objects contained in a single
MATLAB .mat file called "NanoToxRandomForestModels.mat"

MATLAB 2010a is the version utilized to create these models.
They have also been tested with MATLAB version 2012a. 

The designations, input parameters, range of inputs, and 
outputs are described below:

All model outputs describe the predicted measurement taken 
in bronchoalveolar lavage (BAL) fluid from a rodent (rat or
mouse) exposed to the nanoparticles by inhalation, 
intrtracheal instillation, or aspiration. The units of the 
model outputs are all in "fold of control" or the multiple
of the measured response over that of the control group. 

All RF models are only valid within the ranges of specified
input parameters. These models do not extrapolate. They will
produce a prediction outside of their defined ranges, but 
the predicted value will be identical to that of the 
nearest boundary.

------------------------------------------------------------
Input descriptions follow this format:
      Input Description (units) [Minimum - Maximum]

RF_CNT_PMN
   INPUTS AND VALID RANGES: 
      Total Dose, mass (ug/kg) [0 - 6,291]
      Post Exposure, Recovery (days) [1 - 90]
      Median Diameter (nm) [1 - 49]
      Median Length (nm) [320 - 5,900]
      Dose Cobalt (ug/kg) [0 - 3,335]
      Aggregation, MMAD* (nm) [1,670 - 4,200]
   OUTPUTS: PMN (fold of control) -- the multiple change in 
      PMN counts from the control group to the exposed group
      PMN is polymorphonuclear neutrophils. For carbon nano-
      tubes.

RF_CNT_LDH
   INPUTS AND VALID RANGES: 
      Total Dose, mass (ug/kg) [0 - 8,889]
      Post Exposure, Recovery (days) [1 - 90]
      Median Diameter (nm) [1 - 49]
      Median Length (nm) [320 - 5,900]
      Dose Cobalt (ug/kg) [0 - 25,000]
      Aggregation, MMAD* (nm) [1,670 - 4,200]
   OUTPUTS: LDH (fold of control) -- the multiple change in 
      LDH concentration from the control group to the exp-
      osed group. PMN is lactate dehydrogenase. For carbon 
      nanotubes.

RF_TiO2_LDH
   INPUTS AND VALID RANGES:
      Total Dose, mass (ug/kg) [0 - 3.87E6]
      Post Exposure, Recovery (days) [0 - 2]
      Avg. Primary Particle Size (nm) [3.5 - 1,000]
      Aggregation, MMAD (nm) [18 - 1,400]
      Purity (%) [88 - 100]
   OUTPUTS: LDH (fold of control) -- the multiple change in 
      LDH concentration from the control group to the exp-
      osed group. PMN is lactate dehydrogenase. For titanium
      dioxide nanoparticles.

RF_TiO2_TP
   INPUTS AND VALID RANGES:
      Total Dose, mass (ug/kg) [0 - 3.87E6]
      Post Exposure, Recovery (days) [0 - 2]
      Avg. Primary Particle Size (nm) [3.5 - 1,000]
      Aggregation, MMAD (nm) [18 - 1,400]
      Purity (%) [88 - 100]
   OUTPUTS: Total Protein (fold of control) -- the multiple
      change in total protein concentration from the control
      group to the exposure group. For titanium dioxide 
      nanoparticles.

RF_MetOx_LDH
   INPUTS AND VALID RANGES:
      Total Dose, mass (ug/kg) [0 - 16,543]
      Post Exposure, Recovery (days) [1 - 90]
      Aggregation, MMAD (nm) [2,800 - 3,300]
      Purity (%) [90 - 100]
      Gibbs Free Energy (kJ/mol) [-856 - -321]
      Avg. Primary Particle Size (nm) [90 - 452]
   OUTPUTS: LDH (fold of control) -- the multiple change in 
      LDH concentration from the control group to the exp-
      osed group. PMN is lactate dehydrogenase. For metal
      oxide nanoparticles including titanium dioxide, mag-
      nesium oxide, zinc oxide, and silicon dioxide. 

*MMAD is Mass Mode Aerodynamic Diameter

------------------------------------------------------------

To utilize these models, the MATLAB function "predict" comb-
ined with a matrix of the input parameters should be imple-
mented as follows (the order of x1, x2, ... inputs must 
exactly follow the number and order outlined above):

>> y = predict(RF_CNT_PMN,[x1 x2 x3 x4 x5 x6]);

To generate a prediction when any of the input variables is
missing, use the "NaN" (MATLAB designation for 'Not a Num-
ber') in place of the particular value. If NaN is used for
all input parameters, the RF model object will return the
overall mean response for all exposed groups.

------------------------------------------------------------

Further information on these models as well as results can 
be found at:

Gernand J. and Casman E. "Selecting Nanoparticle Properties 
to Mitigate Risks to Workers and the Public – A Machine 
Learning Modeling Framework to Compare Pulmonary Toxicity 
Risks of Nanomaterials." Proc. of IMECE2013. No. 62687.

------------------------------------------------------------