Optimization of a Deep-Learning Method Based on the Classification of Images Generated by Parameterized Deep Snap a Novel Molecular-Image-Input Technique for Quantitative Structure–Activity Relationship (QSAR) Analysis

Abstract

Numerous chemical compounds are distributed around the world and may affect the homeostasis of the endocrine system by disrupting the normal functions of hormone receptors. Although the risks associated with these compounds have been evaluated by acute toxicity testing in mammalian models, the chronic toxicity of many chemicals remains due to high cost of the compounds and the testing, etc. However, computational approaches may be promising alternatives and reduce these evaluations. Recently, deep learning (DL) has been shown to be promising prediction models with high accuracy for recognition of images, speech, signals, and videos since it greatly benefits from large datasets. Recently, a novel DL-based technique called DeepSnap was developed to conduct QSAR analysis using three-dimensional images of chemical structures. It can be used to predict the potential toxicity of many different chemicals to various receptors without extraction of descriptors. DeepSnap has been shown to have a very high capacity in tests using Tox21 quantitative qHTP datasets. Numerous parameters must be adjusted to use the DeepSnap method but they have not been optimized. In this study, the effects of these parameters on the performance of the DL prediction model were evaluated in terms of the loss in validation as an indicator for evaluating the performance of the DL using the toxicity information in the Tox21 qHTP database. The relations of the parameters of DeepSnap such as (1) number of molecules per SDF split into (2) zoom factor percentage, (3) atom size for van der waals percentage, (4) bond radius, (5) minimum bond distance, and (6) bond tolerance, with the validation loss following quadratic function curves, which suggests that optimal thresholds exist to attain the best performance with these prediction models. Using the parameter values set with the best performance, the prediction model of chemical compounds for CAR agonist was built using 64 images, at 105° angle, with AUC of 0.791. Thus, based on these parameters, the proposed DeepSnap-DL approach will be highly reliable and beneficial to establish models to assess the risk associated with various chemicals.

Prediction Model of Aryl Hydrocarbon Receptor Activation by a Novel QSAR Approach, DeepSnap–Deep Learning

“…We have previously shown that hyperparameters in the DeepSnap-DL process affect the performance of prediction models [47][48][49]. Therefore, three main hyperparameters-solver types (STs), batch sizes (BTs), and learning rates (LRs)-in the DeepSnap-DL process were preliminarily optimized using the input data of 201 chemical compound structures.…”

“…By using the resulting image data as input, a prediction model can be classified and constructed by deep learning (DL); thus, we refer to the method as DeepSnap-DL. In addition, we recently reported that the high performance of prediction models of molecular initiating event (MIE) activity for the AOP can be constructed by optimizing hyperparameters and adjusting input data preparation [47][48][49].…”

Deep Learning for Deep Chemistry: Optimizing the Prediction of Chemical Patterns

“…The understanding of chemical systems, and the respective underlying behavior, mechanisms and dynamics, is currently facilitated by the development of descriptive, interpretative, and predictive models, i.e., approximations that represent the target system or process. Applications of such models have included the (i) optimization of reaction parameters and process conditions, e.g., changing the type of reagents, catalysts, and solvents, and also varying systematically, concentration, addition rate, time, temperature, or solvent polarity, (ii) suggestion of new reactions based on critical functional groups, (iii) prediction of reaction/catalyst design, and optimization of heterogeneous/homogeneous catalytic reactions, (iv) acceleration and discovery of new process strategies for batch reactions, (v) establishment of trade-offs in the reaction rate and yield of organic compounds, (vi) description and maximization of the production rate and conversion efficiency of chemical reactions, (vii) prediction of the potential toxicity of different compounds, and also the (viii) rational design of target molecules and guided exploration of chemical space (Kowalik et al, 2012;Zielinski et al, 2017;Häse et al, 2018;Min et al, 2018;Zhou et al, 2018;Ahn et al, 2019;Choi et al, 2019;Gromski et al, 2019;Matsuzaka and Uesawa, 2019).…”

“…Deep learning (DL) approaches can also be particularly useful to solving a variety of chemical problems, including compound identification and classification, and description of soft matter behavior (Huang et al, 2018;Jha et al, 2018;Jørgensen et al, 2018b;Popova et al, 2018;Segler et al, 2018;Zhou et al, 2018;Chandrasekaran et al, 2019;Degiacomi, 2019;Elton et al, 2019;Ghosh et al, 2019;Mater and Coote, 2019;Matsuzaka and Uesawa, 2019;Xu et al, 2019).…”

Tap water fingerprinting using a convolutional neural network built from images of the coffee-ring effect

“…21,22 For example, CNN models have proven the ability to identify brain tumors in magnetic resonance images (MRI) faster and more accurately than the state of the art tools and can identify the pancreas in computerized tomography (CT) images, both of which are challenging analysis problems because of anatomical variability. 10,11 In chemistry, CNN models are being trained using 2D and 3D images of molecular structure for quantitative structure-activity relationship (QSAR) modeling to predict toxicity 23 and to predict therapeutic use classes of drugs. 24 CNN models have also been trained to assign surface-enhanced Raman spectroscopy (SERS) spectra to classes of metabolites and to assign bundles of SERS spectra (8 × 8 pixel hyperspectral images) to the concentration of rhodamine 800 dye at femtomolar concentrations for single molecule detection.…”