Exploring Deep Learning for Metalloporphyrins: Databases, Molecular Representations, and Model Architectures

Abstract

Metalloporphyrins have been studied as biomimetic catalysts for more than 120 years and have accumulated a large amount of data, which provides a solid foundation for deep learning to discover chemical trends and structure-function relationships. In this study, key components of deep learning of metalloporphyrins, including databases, molecular representations, and model architectures, were systematically investigated. A protocol to construct canonical SMILES for metalloporphyrins was proposed, which was then used to represent the two-dimensional structures of over 10,000 metalloporphyrins in an existing computational database. Subsequently, several state-of-the-art chemical deep learning models, including graph neural network-based models and natural language processing-based models, were employed to predict the energy gaps of metalloporphyrins. Two models showed satisfactory predictive performance (R2>0.94) with canonical SMILES as the only source of structural information. In addition, an unsupervised visualization algorithm was used to interpret the molecular features learned by the deep learning models.

Data-driven design of porphyrin photocatalysts using deep transfer learning from quantum chemical databases

“…For a more detailed description of the PBDD, see Table S1 in the Supporting Information (S.I.). In this study, the PBDD was read and preprocessed for deep learning following the protocol proposed in our previous study 32 PorphyBERT. The Bidirectional Encoder Representations from Transformers (BERT) consists of an unsupervised pretraining stage followed by a supervised fine-tuning phase 50 .…”

“…It was first adapted by Schwaller et al as the rxnfp framework (https://rxn4chemistry.github.io/rxnfp/) for the classification of chemical reactions 20 and prediction of chemical reaction yields 51 . Our group further refined the model for multitask property prediction of molecular complexes 27 and energy gap prediction of porphyrin dyes 32 . In this study, we combined features of both models 27,32 into PorphyBERT by first pre-training the model using the MpP structures of PBDD.…”

“…Our group further refined the model for multitask property prediction of molecular complexes 27 and energy gap prediction of porphyrin dyes 32 . In this study, we combined features of both models 27,32 into PorphyBERT by first pre-training the model using the MpP structures of PBDD.…”

“…One of the largest and currently available porphyrin databases is the porphyrin-based dye database (PBDD) 30,31 . Our group recently developed a framework to convert the intrinsically non-generic molecular representation of the PBBD to SMILES, a universal molecular representation for chemical deep learning 32 . With the help of this framework, the PBDD was used to train deep learning models to predict the energy gap of porphyrin dyes.…”