Automatic Selection And Prediction tools for materials and molecules
Project description
ASAP
Automatic Selection And Prediction tools for materials and molecules
Documentation (in progress)
Mapping Materials and Molecules (Paper link)
Cheng B, Griffiths RR, Wengert S, Kunkel C, Stenczel T, Zhu B, Deringer VL, Bernstein N, Margraf JT, Reuter K, Csanyi G. Mapping Materials and Molecules. Accounts of Chemical Research. 2020 Aug 14:12697-705.
This tutorial style paper covers:
-
A concise summary of the theory of representing chemical environments, an a simple yet practical conceptual approach for generating structure maps in a generic and automated manner.
-
Several illustrative examples on mapping material and chemical datasets, including crystalline and amorphous materials, interfaces, and organic molecules. The datasets of these examples are in this repo.
-
Snippets of
asap
commands that were used to analyze the examples and make figures.
Basic usage
Type asap
and use the sub-commands for various tasks.
To get help string:
asap --help
.or. asap subcommand --help
.or. asap subcommand subcommand --help
depending which level of help you are interested in.
-
asap gen_desc
: generate global or atomic descriptors based on the input ASE) xyze file. -
asap map
: make 2D plots using the specified design matrix. Currently PCApca
, sparsified kernel PCAskpca
, UMAPumap
, and t-SNEtsne
are implemented. -
asap cluster
: perform density based clustering. Currently supports DBSCANdbscan
and Fast search of density peaksfdb
. -
asap fit
: fast fit ridge regressionridge
or sparsified kernel ridge regression modelkernelridge
based on the input design matrix and labels. -
asap kde
: quick kernel density estimation on the design matrix. Several versions of kde available. -
asap select
: select a subset of frames using sparsification algorithms.
Quick & basic example
Step 1: generate a design matrix
The first step for a machine-learning analysis or visualization is to generate a "design matrix" made from either global descriptors or atomic descriptors. To do this, we supply asap gen_desc
with an input file that contains the atomic coordintes. Many formats are supported; anything can be read using ase.io is supported. You can use a wildcard to specify the list of input files that matches the pattern (e.g. POSCAR*
, H*
, or *.cif
). However, it is most robust if you use an extended xyz file format (units in angstrom, additional info and cell size in the comment line).
As a quick example, in the folder ./tests/
to generate SOAP descriptors:
asap gen_desc --fxyz small_molecules-1000.xyz soap
for columb matrix:
asap gen_desc -f small_molecules-1000.xyz --no-periodic cm
Step 2: generate a low-dimensional map
After generating the descriptors, one can make a two-dimensional map (asap map
), or regression model (asap fit
), or clustering (asap cluster
), or select a subset of frames (asap select
), or do a clustering analysis (asap cluster
), or estimate the probablity of observing each sample (asap kde
).
For instance, to make a pca map:
asap map -f small_molecules-SOAP.xyz -dm '[SOAP-n4-l3-c1.9-g0.23]' -c dft_formation_energy_per_atom_in_eV pca
You can specify a list of descriptor vectors to include in the design matrix, e.g. '[SOAP-n4-l3-c1.9-g0.23, SOAP-n8-l3-c5.0-g0.3]'
one can use a wildcard to specify the name of all the descriptors to use for the design matrix, e.g.
asap map -f small_molecules-SOAP.xyz -dm '[SOAP*]' -c dft_formation_energy_per_atom_in_eV pca
or even
asap map -f small_molecules-SOAP.xyz -dm '[*]' -c dft_formation_energy_per_atom_in_eV pca
Step 2+: interactive visualization
Using asap map
, a png figure is generated. In addition, the code also output the low-dimensional coordinates of the structures and/or atomic environments. The default output is extended xyz file. One can also specify a different output format using --output
or -o
flag. and the available options are xyz
, matrix
and chemiscope
.
-
If one select
chemiscope
format, a*.json.gz
file will be writen, which can be directly used as the input of chemiscope -
If the output is in
xyz
format, it can be visualized interactively using projection_viewer.
Installation & requirements
python 3
Installation:
python3 setup.py install --user
This should automatically install any depedencies.
List of requirements:
- numpy scipy scikit-learn json ase dscribe umap-learn PyYAML click
Add-Ons:
- (for finding symmetries of crystals) spglib
- (for annotation without overlaps) adjustText
- The FCHL19 representation requires code from the development brach of the QML package. Instructions on how to install the QML package can be found on https://www.qmlcode.org/installation.html.
Additional tools
In the directory ./scripts/ you can find a selection of other python tools.
Tab completion
Tab completion can be enabled by sourcing the asap_completion.sh
script in the ./scripts/ directory.
If a conda environment is used, you can copy this file to $CONDA_PREFIX/etc/conda/activate.d/
to automatically load the completion upon environment activation.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.