Graph Attention Network Model with Defined Applicability Domains for Screening PBT Chemicals

Abstract

In silico models for screening environmentally persistent, bio-accumulative, and toxic (PBT) substances are necessary for sound management of chemicals. Due to the complex structure–activity landscapes (SALs) on the PBT attributes, previous models for screening PBT chemicals lack either applicability domain (AD) characterizations or interpretability, restricting their applications. Herein, graph attention networks (GATs), a novel neural network architecture, were introduced to construct models for screening PBT chemicals. Results show that the GAT model not only outperformed those in previous studies but also exhibited interpretability since it optimizes attention weight parameters (P AW) that indicate contributions of each atom to the PBT attributes. An AD characterization termed ADFP–AC, which considers both molecular fingerprint (FP) similarities and compounds at activity cliffs (ACs) of SALs, was proposed to describe the ADs, which further assured the performance of the GAT model. Eight previously unidentified classes of compounds were identified as PBT chemicals from the Inventory of Existing Chemical Substances in China. The GAT model together with the ADFP–AC characterization may serve as efficient tools for screening PBT chemicals, and the modeling methodology can be applied to other physicochemical, environmental, behavioral, and toxicological parameters of chemicals that are necessary for their risk assessment and management.

Machine Learning Model for Screening Thyroid Stimulating Hormone Receptor Agonists Based on Updated Datasets and Improved Applicability Domain Metrics

“…AD Characterization. Based on previous research gains 19,20 and SAL analysis, an AD characterization method abbreviated as AD SAL {ρ s , I A } was raised in the current study, where ρ s stands for molecular similarity density and I A describes inconsistency in molecular activities or molecular local discontinuity. ρ s defines weighted similarity density between a query compound and training compounds:…”

“…Recently, an AD characterization (AD FP-AC ), where the subscript "FP" represents molecular fingerprints and "AC" represents ACs in SALs, was proposed. 19,20 The AD FP-AC employs local discontinuity scores (S D ) to identify compounds on the ACs and was proven to outperform a previous AD characterization method for ML-based QSARs. 20 The AD FP-AC calculates molecular similarity (S M ), employs a cutoff value of S M (S cutoff ) to determine neighboring training compounds of a query compound, and employs the neighboring training compounds to calculate arithmetically the average S M and S D .…”

“…19,20 The AD FP-AC employs local discontinuity scores (S D ) to identify compounds on the ACs and was proven to outperform a previous AD characterization method for ML-based QSARs. 20 The AD FP-AC calculates molecular similarity (S M ), employs a cutoff value of S M (S cutoff ) to determine neighboring training compounds of a query compound, and employs the neighboring training compounds to calculate arithmetically the average S M and S D . Nevertheless, the AD characterization can be improved with refined calculation on S M .…”

Machine Learning-Based Models with High Accuracy and Broad Applicability Domains for Screening PMT/vPvM Substances

“…13 Thus, machine learning (ML) has attracted attention for the rapid and accurate prediction of substance properties crucial to the identification of PMT/ vPvM substances because of its versatility. 14,15 A key to the success of ML modeling is the need for diverse datasets and appropriate algorithms, which directly influence applicability domains (ADs) and generalization ability. 16,17 Limited data for ML model construction may lead to overfitting, narrow ADs, and poor generalization, whereas inadequate ML algorithms only serve as a black box without any interpretation value to reveal the mechanism that relates imported information to the modeling results.…”

“…However, there is a pressing demand for low-cost and high-throughput protocols for screening PMT/vPvM substances, as the testing process for acquiring these parameters requires numerous expensive and time-consuming experimental protocols . Thus, machine learning (ML) has attracted attention for the rapid and accurate prediction of substance properties crucial to the identification of PMT/vPvM substances because of its versatility. , A key to the success of ML modeling is the need for diverse datasets and appropriate algorithms, which directly influence applicability domains (ADs) and generalization ability. , Limited data for ML model construction may lead to overfitting, narrow ADs, and poor generalization, whereas inadequate ML algorithms only serve as a black box without any interpretation value to reveal the mechanism that relates imported information to the modeling results. − ML models built upon big data and multi-algorithms are thus urgently needed as a tool for the rapid and accurate identification of PMT/vPvM substances.…”