Day 2B - Geometric Morphometrics in R PDF

G562 Geometric Morphometrics
A basic geometric morphometric analysis

Step by step in R
Department of Geological Sciences | Indiana University
(c) 2012, P. David Polly
R packages to install
geomorph shapes svd scatterplot3d rgl MASS ape vegan Geometric morphometrics package by Adams and Otrola-Castillo Geometric morphometrics package by Ian Dryden Singular value decomposition package Functions for 3D plotting (installed as dependency to above) More 3D functions (installed as dependency to above) Modern Applied Statistics with S (installed as dependency to above) Analyses of Phylogenetics and Evolution (installed as dependency to above) Community ecology package (installed as dependency to above)
Example file: Bialowieza_Bialowieza.tps Install from CRAN Load with library(packagename)

General Procedure
Study design Data collection Data standardization Analysis Results interpretation
Steps in a geometric morphometric analysis

1. Obtain landmark coordinates. 2. Standarization: Procrustes superimposition of the landmarks. removes size, translation, and rotation (also project to tangent space for further statistical analysis). 3. Standardization: Create shape variables (variables that retain information about shape, but are uncorrelated with one another and have appropriate degrees of freedom) PCA scores are one kind of shape variable that satisfy these requirements 4. Analysis: Conduct further analyses on the shape variables Type of analysis required depends on questions being addressed Regression, ANOVA, path analysis, tree-building, etc.
Choosing landmarks
1. landmarks should sample the aspects of the shape that are of interest 2. enough landmarks should be used to adequately sample the shape 3. dont oversample (NB: each landmark adds weight to the
analysis. Multiple landmarks on one area will increase weight of that area.)
4. landmarks should be repeatable (same point on every specimen, placed with as little error as possible)
1. type 1 = location of the point defined by obvious biologically homologous structures (eg. intersection of three bones) 2. type 2 = location of the point defined by obvious geometry (eg., point of greatest curvature) 3. type 3 = location of point defined with reference to another point (eg., point ventral to last tooth)
5. landmarks must be placed in the same order on all specimens
Obtaining landmark coordinates

From a file lands <- readland.tps(file.choose()) lands <- readland.nts(file.choose()) [imports from TPS format file] [imports from NTS format file]
From an image digitize2d(filename, landmarks, scale) [opens jpg file and collects lands]
Note: digitize2d may need debugging. Saves coordinates to file in working directory in NTS format.
readmulti.nts(filenames)

digitize2d()
Erik Otarola-Castillo has sent a fix. Do the following and the function will work: 1. Set the working directory to where your images are stored. 2. Define the function picscale() as follows
picscale <- function(scale){ digscale<-NULL digscale<-locator(2,type="o",lwd=2,col="red",lty="11") sqrt(sum(diff(digscale$x)^2 + diff(digscale$y)^2))*scale }
To collection landmarks in digitize2d:

1. If you have a scale bar in the images
1.1. enter the length of the scale bar as argument (e.g., 10 if the scale bar is 10 mm) 1.2. when digitize2d starts, click on both ends of the scale bar, then click on your landmarks in the proper order 1.3. coordinates and centroid sizes will be scaled in the units of your scale bar (e.g., mm)
2. If you dont have a scale bar:

2.1. enter 1 as the scale argument 2.2. when digitize2d starts, click in two arbitrary but different places on the image, then click on your landmarks in the proper order 2.3. coordinates will have no real units and centroid size will be meaningless
Automate the digitizing and loading of files

Digitize.My.Files <- edit()
function(path, lands, scale) { setwd(path) myFiles <- dir(pattern="[jJ][pP][gG]") for( i in 1:length( myFiles ) ) { dig2d(myFiles[i],lands, scale) } myFiles <- dir(pattern="[nN][tT][sS]") return(readmulti.nts(myFiles)) }
mylands <- Digitize.My.Files(/Users/pdavidpolly/ShrewsAndMarmots, 10, 10)
Cryptic error?
Error in .External2(C_edit, name, file, title, editor) : unexpected input occurred on line 3 use a command like x <- edit() to recover
May be caused by smart quotes if you copied and pasted script from Word or other text editor
Performing Procrustes analysis and obtaining shape variables

Also known as Generalized Procrustes Analysis (GPA) > gpa.lands <- gpagen(lands) results of gpagen() include:
1. a plot of the superimposed specimens and their consensus (mean) shape 2. superimposed coordinates ($coords) 3. shape variables as principal components scores ($pc.scores)
Procrustes of shrew teeth

data from Bialowieza_Bialowieza.tps
Before
After
> plotAllSpecimens(lands)
> gpagen(lands)
Details of gpagen()
proc <- gpagen(
This function does a generalized Procrustes analysis, superimposition of multiple specimens about their mean (Gower, 1975; Rohlf and Slice, 1990) Returns a plot of superimposed coordinates Returns the superimposed coordinates and their centroid sizes in dataframe format:
$coords = the x,y(,z) Procrustes coordinates after superimposition $Csize = the centroid size of the specimens
To plot the Procrustes coordinates yourself:

for(i in 1:dim(proc$coords)[3]) points(proc$coords[,,i]) plotAllSpecimens(proc$coords)
Creating a principal components plot

> pca.lands <- plotTangentSpace(proc$coords, label=TRUE) results of plotTangentSpace() include: 1. a plot of the specimens in principal components space and two thin plate spline grids showing the shapes associated with the positive and negative ends of the horizontal axis (default = PC1 x PC2 set axes using the args axis1 and axis2) 2. summary of variance associated with each PC axis ($pc.summary) 3. shape variables as principal components scores ($pc.scores)
Principal component plot of shrew teeth

Thin plate spline grid of shape at negative end of 1st axis
PC 2
Thin plate spline grid of shape at positive end of 1st axis PC 1

Variance explained by PC axes

pca.lands$pc.summary
Important concept: variance of a data set

Variance = variability = shape variation PC scores preserve the shape variation in the original data Variance = average squared distance from the mean Variance in Procrustes coordinates = Variance in PC scores
> sum(pca.lands$pc.summary$sdev^2) > sum(apply(pca.lands$pc.scores, 2, var)) > sum(apply(proc$coords,c(1,2),var))
Show scree plot of variance explained

barplot(pca.lands$pc.summary$sdev^2/sum(pca.lands$pc.summary$sdev^2))
Create your own customized plot

> plot(pca.lands$pc.scores[,1:2],pch=15,xlab="PC1",ylab="PC2") > text(pca.lands$pc.scores[,1:2],rownames(pca.lands$pc.scores[,1:2]),pos=4,cex=.5)
Create 3D scatterplot
> library(scatterplot3d) > scatterplot3d(pca.lands$pc.scores[,1:3])
Terminology: shape
Shape = a set of landmarks Sometimes shape specifically refers to the landmark configurations without respect to size and form refers to the configurations including size
Terminology: Consensus shape

Consensus shape = mean shape Average x,y coordinate of each landmark after Procrustes alignment
To calculate
> consensus <- apply(proc$coords, c(1,2), mean)
> plot(consensus,asp=1, type="n") > for(i in 1:length(proc$coords[,,3])) points(proc$coords[,,i]) > points(consensus, col=Red, cex=2, pch=20)
Terminology: centroid
The mean (center) of a shape (or of a landmark). Average x,y of all coordinates across all landmarks. (for individual landmark, the centroid is the consensus point for that landmark). To calculate:
> centroid <- apply(proc$coords,2,mean)
Finding your way around shape space

150
Scores on PC1 and PC2 for first specimen

100
Position of consensus shape

[-67.3,0.77]
Squirrel Squirrel Mouse Shrew Shrew
PC2
50
Scores on PC1 and PC2 for fourth specimen

-150 -100
-50
[40.2,-22.2]
-150
-100
-50
0 PC1
50
100
150
Visualizing shapes: thin plate spline grids

Allows difference between two shapes to be compared as a grid of first being deformed into second. Usually differences are shown between real (or imaginary) specimens and the consensus shape.
> plotRefToTarget(consensus,proc $coords[,,1])
Principles of shape space

1. PCA space = shape space 2. Every point in shape space corresponds to configuration of landmarks 3. By definition, the consensus (mean shape) lies at the origin (0,0) of the shape space 4. PC scores are the coordinates of the PCA plot (they are the addresses of the points in the shape space)
Landmark Space (specimen space)

Two (or three) dimensional space (x, y, z) Numbers associated with space are the landmark coordinates Many points per specimen
Shape Space (PCA space)

Multidimensional space (nearly as many dimensions as there are landmarks x coordinates) Numbers are scores (addresses in PC space) One point per specimen
Basic steps of shape analysis in R

Capture images digital camera, etc. Collect coordinates digitize2d(filename) Read coordinate files in NTS format readmulti.nts(filenames) Superimpose landmarks and do PCA gpagen(lands) View PCA results plotTangentSpace($coords) Look at shape differences plotRefToTarget(shape1, shape2)
PCA of faces
Axis 2 -0.05 0.00
0.05
-0.10
-0.05 Axis 1
0.00
0.05
0.10
Daniel2 Derek1 DavidP1 0.05 Silvia1
PC2
Andrew2 Thomasz2 Aidan1 Silvia2 DavidP2 Nadia1 DavidW2 Julietta1 Nadia2 DavidW1 Mark1 Julietta2 Neil2 Emily2Neil1 Emily1 Daniel1 Mark2 Derek2 Thomasz1 Stephanie1 Aidan2 Stephanie2
0.00
Andrew1 -0.05 -0.15
-0.10
-0.05 PC1
0.00
0.05
0.10
0.10
Daniel2 Derek1 DavidP1 0.05 Silvia1
PC2
Andrew2 Thomasz2 Aidan1 Silvia2 DavidP2 Nadia1 DavidW2 Julietta1 Nadia2 DavidW1 Mark1 Julietta2 Neil2 Emily2Neil1 Emily1 Daniel1 Mark2 Derek2 Thomasz1 Stephanie1 Aidan2 Stephanie2
0.00
Andrew1 -0.05 -0.15
-0.10
-0.05 PC1
0.00
0.05
0.10

Day 2B - Geometric Morphometrics in R PDF

Uploaded by

Copyright:

Available Formats

Day 2B - Geometric Morphometrics in R PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Day 2B - Geometric Morphometrics in R PDF

Uploaded by

Copyright:

Available Formats

G562 Geometric Morphometrics

A basic geometric morphometric analysis

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Example file: Bialowieza_Bialowieza.tps Install from CRAN Load with library(packagename)

G562 Geometric Morphometrics

G562 Geometric Morphometrics

Steps in a geometric morphometric analysis

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

5. landmarks must be placed in the same order on all specimens

G562 Geometric Morphometrics

Obtaining landmark coordinates

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

Department of Geological Sciences | Indiana University

To collection landmarks in digitize2d:

2. If you dont have a scale bar:

G562 Geometric Morphometrics

Automate the digitizing and loading of files

mylands <- Digitize.My.Files(/Users/pdavidpolly/ShrewsAndMarmots, 10, 10)

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Performing Procrustes analysis and obtaining shape variables

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Procrustes of shrew teeth

(c) 2012, P. David Polly

G562 Geometric Morphometrics

proc <- gpagen(

To plot the Procrustes coordinates yourself:

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Creating a principal components plot

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Principal component plot of shrew teeth

Thin plate spline grid of shape at positive end of 1st axis PC 1

G562 Geometric Morphometrics

Variance explained by PC axes

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Important concept: variance of a data set

Department of Geological Sciences | Indiana University

(c) 2012, P. David Polly

G562 Geometric Morphometrics

Show scree plot of variance explained

Department of Geological Sciences | Indiana University