auto3dgm-matlab-gorgon

MST-based Generalized Dataset Procrustes Distance by Jesús Puente

MATLAB Code originally written by Jesús Puente ([email protected]); forks of this code include PuenteAlignment, maintained by Tingran Gao ([email protected]), and auto3dgm-matlab-gorgon, maintained by Julie Winchester ([email protected]). This code has also been ported to R by Christopher Glynn ([email protected]) under the name auto3dgm.

auto3dgm-matlab-gorgon is a fork of auto3dgm with a focus on providing improved user experience and additional functionality both upstream and downstream frmo the core workflow. The latest commits can be found in the dev branch, but users are encouraged to clone the master branch as it contains stable milestone releases. PuenteAlignment and auto3dgm-matlab-gorgon both have the same core MST-based alignment functionality. Auto3dgm-matlab-gorgon does not currently include the two spectral relaxation alignment algorithms included in the current PuenteAlignment version. Auto3dgm-matlab-gorgon does implement two new features currently: user control over how overall mesh alignment is handled after all sample meshes are aligned to each other, and principal components analysis of partial procrustes tangent shape coordinates. These new features are described below. This branch also has a number of smaller user experience improvements, such as support for OSX local analyses or more flexible directory formatting in jadd_path.m.

Overall mesh alignment

Users can control the overall alignment of surface meshes (after meshes are aligned to each other) by setting the 'align_to' field in jadd_path.m. If this value is set to the name of a surface mesh file in the current analysis, then all meshes will be aligned to that mesh's orientation in 3D coordinate space. If 'align_to' is set to 'auto' or blank (''), auto3dgm will attempt to derive a semi-standardized overall alignment for all meshes. Principal axes of shape variation are derived for all sample vertex point clouds concatenated, and each mesh is rotated so that first, second, and third principal axes of shape variation correspond to Y, X, and Z coordinate axes respectively.

PCA of partial procrustes tangent shape coordinates

If the 'do_tangent_pca' field in jadd_path.m is set to 1, then auto3dgm carries out a principal components analysis of partial procrustes tangent shape coordinates. The method used for the tangent space projection is that of Dryden and Mardia's (1993, 1998, 2016) paper and textbook for unit-scaled shape data, but using the calculations from Rohlf's (1999) description of 'Kendall tangent space coordinates'. Several files are saved with PCA results, including a basic scatter plot of PC1 and PC2, PC scores per individual, eigenvalues and percent variance explained per PC, and eigenvectors per PC.

---

Using auto3dgm-matlab-gorgon

The current version of auto3dgm-matlab-gorgon supports standard sequential execution on any matlab installation or parallel computation on a cluster managed by Sun Grid Engine (SGE). Follow the instructions below for installation and usage of this software.

1. Get the current version of auto3dgm-matlab-gorgon. Either download the repository files directly to the desired path, or if git is installed simply cd into your desired path, then type

    git clone https://github.com/JuliaWinchester/auto3dgm-matlab-gorgon.git
   

2. Find the script jadd_path.m in the folder auto3dgm-matlab-gorgon/code/, and set paths and parameters there. If you are using parallel computation and assign an email address to the variable email_notification, a notification will be sent automatically to that email address whenever a cluster job completes or aborts.

3. Launch MATLAB and cd into the folder auto3dgm-matlab-gorgon/code/. If sequential non-cluster analysis is desired, type main to run the script main.m. If parallel cluster analysis is desired, type clusterMain to run the script clusterMain.m. Jobs should be submitted to the cluster subsequently. Use qstat to monitor job status. Analysis is complete when job clusterRun is finished.

The script clusterRun.m automates the four major steps of auto3dgm parallel computation. It is also possible to run these four steps manually following these instructions.

1. Launch MATLAB, cd into the folder auto3dgm-matlab-gorgon/code/, and type clusterMapLowRes. Low-resolution alignment jobs should then be submitted to the cluster. Use qstat to monitor job status.

2. After all jobs are completed, type clusterReduceLowRes. This generates low-resolution alignment results in the output folder specified in jadd_path.m.

3. Type clusterMapHighRes to submit high-resolution alignment jobs to the cluster.

4. After all jobs are completed, type in clusterReduceHighRes. This generates high-resolution alignment results in the output folder specified in jadd_path.m.

---

WebGL-based Alignment Visualization

After the alignment process is completed, the result can be visualized using a javascript-based viewer located under the folder viewer/. See here for an online demo.

1. Move all output files ending with "_aligned.obj" from the subfolder aligned/ (under your output folder) to the subfolder viewer/aligned_meshes/.

2. Set up an HTTP server under the folder viewer/. (If you already placed the folder viewer/ somewhere with HTTP services, feel free to skip this step.) For instance, you can cd viewer/ and type into the terminal

    python -m SimpleHTTPServer 8000
   

3. Launch your browser and direct it to http://localhost:8000/auto3dgm.html.

---

Please Cite:

Boyer, Doug M., et al. A New Fully Automated Approach for Aligning and Comparing Shapes. The Anatomical Record 298.1 (2015): 249-276.

Puente, Jesús. Distances and Algorithms to Compare Sets of Shapes for Automated Biological Morphometrics. PhD Thesis, Princeton University, 2013.