New tricks by very old dogs: Predicting the catalytic hydrogenation of HMF derivatives using Slater-type orbitals

The models were created using the Orthogonal Partial Least Squares (OPLS) method implemented in Simca-P+ 12 (Umetrics). This method was first introduced by Trygg et al., and is basically an extension of the Partial Least Squares (PLS) approach. It combines Orthogonal Signal Correction (OSC, an algorithm often applied in spectroscopic data analysis) with classical PLS regression. Using OPLS, the first step is orthogonalising the variables and responses, thus establishing a 1:1 relationship between the variables (x) and each response (y). This simplifies the interpretation of the results. When the responses are correlated, this will also show up as a result. We preferred using OPLS in this case over PLS because of the internal correlations in the descriptor matrix. Note, however, that using OPLS does not improve the model’s performance – it only simplifies the interpretation. Figure S1 shows a block flowchart of the OPLS method. A detailed technical description of the OPLS algorithm is given in the work of Trygg and Wold.