Metal-Oxide-LASSO-lo

Code written by A S M Jonayat Ph.D. Candidate Dept. of Mechanical and Nuclear Engineering The Pennsylvania State University, University Park - 16802

Distributed as part of the publication - Interaction trends between single metal atoms and oxide supports identified with density functional theory and statistical learning Nolan J. O’Connor, A S M Jonayat, Michael J. Janik*, Thomas P. Senftle*

Full Article: www.nature.com/articles/s41929-018-0094-5 Read Only: https://rdcu.be/2kdc

Department of Chemical Engineering, Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, University Park, PA 16802 (USA) Department of Chemical and Biomolecular Engineering, Rice University, Houston, TX 77005 (USA) *mjanik@engr.psu.edu *tsenftle@rice.edu

The Code is devided into three parts -

1. Step 1: Read excel file with primary descriptors and generate secondary dataset
2. Step 2: Use LASSO to find out top descriptors for Binding energy
3. Step 3: Use lo regression to find best 1D-5D descriptors.

All files are provided in this directory.

Step 1: Dir: ./Step_1 MATLAB CODE: cluster.m INPUT: Binding_energyV4.xlsx OUTPUT: data_set.mat (matlab data file)

Step 2: Dir: ./Step_2 MATLAB CODE: analysis.m INPUT: data_set.mat OUTPUT: a. lasso_final.mat (matlab data file) b. lasso_param.txt (results of LASSO analysis)

Step 3: Dir: ./Step_3 MATLAB CODE: analysis.m OUTPUTS: a. log_L1.txt : Results of lo fits b. jpgs for DFT vs predicted for 1D-5D c. raw data for plots in b.