An Ensemble Structure and Physicochemical (SPOC) Descriptor for Machine‐Learning Prediction of Chemical Reaction and Molecular Properties

Abstract

Feature representations, or descriptors, are machines’ chemical language that largely shapes the prediction capability, generalizability and interpretability of machine learning models. To develop a generally applicable descriptor is highly warranted for chemists to deal with conventional prediction tasks in the context of sparsely distributed and small datasets. Inspired by the chemist's vision on molecules, we presented herein an ensemble descriptor, SPOC, curated on the principles of physical organic chemistry that integrates Structure and Physicochemical property (SPOC) of a molecule. SPOC could be readily constructed by combining molecular fingerprints, representing the structure of a given molecule, and molecular physicochemical properties extracted from RDKit or Mordred molecular descriptors. The applicability of SPOC was fully surveyed in a range of well‐structured chemical databases with machine learning tasks varying from regression to classifications.

Molecular Toxicity Virtual Screening Applying a Quantized Computational SNN-Based Framework

“…The first comparison was conducted over the results presented by [ 40 ] using a random forest (i.e., RF) and Morgan MFs on BBBP (0.909 ± 0.028 AUC), Tox21 (0.819 ± 0.017), SIDER (0.687 ± 0.014), and ClinTox (0.759 ± 0.060). In [ 41 ], the authors evaluated MAACS fingerprints over the Tox21 dataset achieving an AUC of 0.805 ± 0.01, an AUC of 0.721 ± 0.004 for BBBP, and an AUC equal to 0.797 ± 0.151 for Clintox, applying an ensemble of decision trees over 5-fold cross-validation. Another paper [ 42 ] focused on the Tox21 dataset reporting the outcomes of the in silico toxicity evaluation by five classifiers on Morgan fingerprints: the LightGBM overperformed other classifiers, reaching an AUC of 0.795 on the test set (standard deviation was not reported) for NR-AR.…”

“…Standard deviation was included if reported in the original papers. AUC values from [ 40 , 41 , 42 , 43 , 44 , 45 , 47 , 48 ].…”

“… Outcomes for BBBP and Clintox compared to the previous literature. AUC values as reported in [ 40 , 41 , 46 , 49 , 51 , 53 , 54 , 55 , 56 ]. …”

Reaxtica: A knowledge-guided machine learning platform for fast and accurate reaction selectivity and yield prediction

“…17 Yang et al combined fingerprint and physicochemical descriptors for reaction prediction. 18 However, these descriptors are not easy to interpret nor generalize well outside the training reaction space except the Hammett and TSEI descriptors.…”

Reaxtica: A knowledge-guided machine learning platform for fast and accurate reaction selectivity and yield prediction

“…17 Luo et al combined fingerprint and physicochemical descriptors for reaction prediction. 18 However, these descriptors are not easy to interpret nor generalize well outside the training reaction space except the Hammett and TSEI descriptors.…”