A general optimization protocol for molecular property prediction using a deep learning network

Abstract

The key to generating the best deep learning model for predicting molecular property is to test and apply various optimization methods. While individual optimization methods from different past works outside the pharmaceutical domain each succeeded in improving the model performance, better improvement may be achieved when specific combinations of these methods and practices are applied. In this work, three high-performance optimization methods in the literature that have been shown to dramatically improve model performance from other fields are used and discussed, eventually resulting in a general procedure for generating optimized CNN models on different properties of molecules. The three techniques are the dynamic batch size strategy for different enumeration ratios of the SMILES representation of compounds, Bayesian optimization for selecting the hyperparameters of a model and feature learning using chemical features obtained by a feedforward neural network, which are concatenated with the learned molecular feature vector. A total of seven different molecular properties (water solubility, lipophilicity, hydration energy, electronic properties, blood–brain barrier permeability and inhibition) are used. We demonstrate how each of the three techniques can affect the model and how the best model can generally benefit from using Bayesian optimization combined with dynamic batch size tuning.

Experimental and Computational Methods to Assess Central Nervous System Penetration of Small Molecules

“…A deep learning (DL) method achieved better accuracy than the ML methods on three different datasets [ 41 ]. Recently, various DL methods such as artificial neural networks (ANN), deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), and graph convolutional neural networks (GCNN) have been used to predict BBB permeability with high accuracy [ 42 , 43 , 44 , 45 , 46 , 47 , 48 ]. Alsenan et al [ 49 ] published a highly accurate deep learning model based on a recurrent neural network.…”

On modeling and utilizing chemical compound information with deep learning technologies: A task-oriented approach

“…FP2VEC [110] achieved AUC-ROC of 0.880 in Tox21 dataset by employing a multi-task learning framework. ConvS2S [126] improved the performances in various datasets including solubility, BBBP, and HIV datasets by suggesting SMILES augmentation scheme.…”

Predicting Binding Energies and Electronic Properties of Boron Nitride Fullerenes Using a Graph Convolutional Network

“…Indeed, the atomic coordinates are not necessary information if the molecule or cluster has a well-defined bonding topology. For instance, a molecule can be described by a simplified molecular-input line-entry system (SMILES) string, which can predict the key chemical and biological properties of the molecule. , Moreover, the DL technique has been utilized for molecular generation, odor identification, and toxicity prediction …”

“…For instance, a molecule can be described by a simplified molecular-input line-entry system (SMILES) string, which can predict the key chemical and biological properties of the molecule. 30,31 Moreover, the DL technique has been utilized for molecular generation, 32 odor identification, 33 and toxicity prediction. 34 Among current DL algorithms, the graph neural network (GNN) represents one of the most rapidly growing classes, and its applications in physical science span from predicting the energy, force, and physical properties of molecules and materials to accelerated atomistic simulations to forecasting reactivity and synthesis routes for molecules and materials.…”