Machine learning models for hydrogen bond donor and acceptor strengths using large and diverse training data generated by first-principles interaction free energies

Abstract

We present machine learning (ML) models for hydrogen bond acceptor (HBA) and hydrogen bond donor (HBD) strengths. Quantum chemical (QC) free energies in solution for 1:1 hydrogen-bonded complex formation to the reference molecules 4-fluorophenol and acetone serve as our target values. Our acceptor and donor databases are the largest on record with 4426 and 1036 data points, respectively. After scanning over radial atomic descriptors and ML methods, our final trained HBA and HBD ML models achieve RMSEs of 3.8 kJ mol−1 (acceptors), and 2.3 kJ mol−1 (donors) on experimental test sets, respectively. This performance is comparable with previous models that are trained on experimental hydrogen bonding free energies, indicating that molecular QC data can serve as substitute for experiment. The potential ramifications thereof could lead to a full replacement of wetlab chemistry for HBA/HBD strength determination by QC. As a possible chemical application of our ML models, we highlight our predicted HBA and HBD strengths as possible descriptors in two case studies on trends in intramolecular hydrogen bonding.

Prediction of Chameleonic Efficiency

“…These methods are computationally expensive, but the implementation of a machine learning model for prediction of hydrogen bond energies based on a large QM dataset was recently reported. [22] We envision that such ML models for hydrogen bonding strengths will make routine assessment of donor and acceptor sites very fast.…”

“…Compounds 1-24 were sourced via Aldrich Market Select using the following suppliers: Adooq Bioscience (California, USA) -boceprevir (13), dabigatran etexilate (3), daclatasvir (19), eliglustat (7), ledipasvir (23), tacrolimus (10), telaprevir (14); Aldrich-CPR (Sigma-Aldrich, Wisconsin, USA) -erythromycin (9); Ambeed Inc (Illinois, USA) -aliskiren (1), atazanavir (4), sofosbuvir (8), venetoclax (11); Carbosynth Ltd (Oxfordshire, UK) -pibrentasvir (21); Cayman Chemical -simeprevir (18), velpatasvir (24); MedChem Express (New Jersey, USA) -avanafil (2), cobicistat (5), elbasvir (22), glecaprevir (15), ombitasvir (20); Target Molecule Corp. (Massachusetts, USA) -asunaprevir (12), edoxaban (6), grazoprevir (16), paritaprevir (17). Stock solutions in DMSO with a concentration of 10 mM were prepared and used for all analyses.…”

Prediction of Chameleonic Efficiency

“…Compounds 1-24 were sourced via Aldrich Market Select using the following suppliers: Adooq Bioscience (California, USA) -boceprevir ( 13), dabigatran etexilate (3), daclatasvir (19), eliglustat (7), ledipasvir (23), tacrolimus (10), telaprevir (14); Aldrich-CPR (Sigma-Aldrich, Wisconsin, USA) -erythromycin (9); Ambeed Inc (Illinois, USA) -aliskiren (1), atazanavir (4), sofosbuvir (8), venetoclax (11); Carbosynth Ltd (Oxfordshire, UK) -pibrentasvir (21); Cayman Chemical -simeprevir (18), velpatasvir (24); MedChem Express (New Jersey, USA) -avanafil (2), cobicistat (5), elbasvir (22), glecaprevir (15), ombitasvir (20); Target Molecule Corp. (Massachusetts, USA) -asunaprevir (12), edoxaban (6), grazoprevir (16), paritaprevir (17). Stock solutions in DMSO with a concentration of 10 mM were prepared and used for all analyses.…”

RegioML: predicting the regioselectivity of electrophilic aromatic substitution reactions using machine learning

“…9 We investigate seven different atomic descriptors of intermediate complexity as input to the ML models (details on the descriptors are given in Table S2 in the supporting information). The atomic descriptors are developed by Finkelmann et al 20,21 and are chosen because they have been successfully applied to the prediction of site of metabolism, 21,22 hydrogen bond donor and acceptor strengths, 23,24 and Ames mutagenicity of primary aromatic amines. 25 Almost all of the descriptors depend on charge model 5 (CM5) atomic charges, 26 which are obtained from a single point calculation using GFN1-xTB as implemented in the open source semiempirical software package xtb version 6.4.0.…”