Tutorial on Generative Topographic Mapping

The tutorial aims at presenting the Generative Topographic Mapping (GTM) algorithm [Bishop et al, Neural Computation 10, No. 1, 215–234 (1998)].

The GTM is an unsupervised method to map high dimensional data to a two-dimensional representation. In the process, the GTM builds a probabilistic model of the data that can be exploited for data characterization, comparison or classification and regression model building. The GTM approach will be used to analyze a dataset of flavors and to explore structure-flavor relationships. It will be the occasion to get some deeper insight into the method with a particular focus on the effects of the GTM parameterization on the obtained map.

Software

The tutorial is based on three pieces of software :

• xGTMapTool : a graphical user interface frontend for the preparation of a GTM.
• xGTMview : an application to link the GTM trained on chemical data and the chemical structures.
• xGTMmanifold : an application illustrating the concept of GTM manifold and data space.

The software are supplied on the USB key and can be downloaded for the OS of your choice :

• Windows
• Mac
• Linux

Updated versions can be requested by contacting F. Bonachera

The licence of the software is distributed freely and a licence file, called « licence.dat » is distributed with the software for the OS of your choice (Windows, Mac or Linux).

The licence file must be installed in a proper location to be found.

• On Windows : create the directory AppData\local\ISIDAGTM2018 directory at the root of your home directory and copy the file license.dat in it. The absolute path of the file should be similar to this one :

C :\Users\username\AppData\local\ISIDAGTM2018\licence.dat

The file and the directory should have read and write permissions.

• On Mac : create the directory .config/ISIDAGTM2018 directory at the root of your home directory and copy the file license.dat in it. The absolute path of the file should be similar to this one :

/Users/username/.config/ISIDAGTM2018/licence.dat

• On Linux : create the directory .config/ISIDAGTM2018 directory at the root of your home directory and copy the file license.dat in it. The absolute path of the file should be similar to this one :

/home/username/.config/ISIDAGTM2018/licence.dat

Datasets

The tutorial uses a dataset of organoleptic compounds mined (in February 2018) from the FlavorDB database [Nucleic Acids Research, 2018, Vol. 46, Database issue].

The data are distributed on the USB key and can be downloaded here : http://infochim.u-strasbg.fr/CS3_2018/Tuto1/Data/FDB.zip

Step-by-step instructions

The protocol of the tutorial session is available in the following document : http://infochim.u-strasbg.fr/CS3_2018/Tuto1/Tutorial_GTM+DH_20_06_2018.pdf