A novel adaptive ensemble classification framework for ADME prediction

Abstract

It has now become clear that in silico prediction of ADME (absorption, distribution, metabolism, and elimination) characteristics is an important component of the drug discovery process. Therefore, there has been considerable interest in the development of in silico modeling of ADME prediction in recent years. Despite the advances in this field, there remains challenges when facing the unbalanced and high dimensionality problems simultaneously. In this work, we introduce a novel adaptive ensemble classification framework named as AECF to deal with the above issues. AECF includes four components which are (1) data balancing, (2) generating individual models, (3) combining individual models, and (4) optimizing the ensemble. We considered five sampling methods, seven base modeling techniques, and ten ensemble rules to build a choice pool. The proper route of constructing predictive models was determined automatically according to the imbalance ratio (IR). With the adaptive characteristics of AECF, it can be used to work on the different kinds of ADME data, and the balanced data is a special case in AECF. We evaluated the performance of our approach using five extensive ADME datasets concerning Caco-2 cell permeability (CacoP), human intestinal absorption (HIA), oral bioavailability (OB), and Pglycoprotein (P-gp) binders (substrates/inhibitors, PS/PI). The performance of AECF was evaluated on two independent datasets, and the average AUC values were 0.8574-0.8602, 0.8968-0.9182, 0.7821-0.7981, 0.8139-0.8311, and 0.8874-0.8898 for CacoP, HIA, OB, PS and PI, respectively. Our results show that AECF can provide better performance and generality compared with individual models and two representative ensemble methods bagging and boosting. Furthermore, the degree of complementarity among the AECF ensemble members was investigated for the purpose of elucidating the potential advantages of our framework. We found that AECF can effectively select complementary members to construct predictive models by our auto-adaptive optimization approach, and the additional diversity in both sample and feature space mainly contribute to the complementarity of ensemble members.

A Study on the Therapeutic Efficacy of San Zi Yang Qin Decoction for Non-Alcoholic Fatty Liver Disease and the Underlying Mechanism Based on Network Pharmacology

“…To estimate their drug-likeness, the physicochemical characteristics of collected chemicals were compared with those of known drugs, according to their AMDE (absorption, metabolism, distribution, excretion). In 2012, Liang et al proposed that quantitative estimate of drug-likeness (QED) [ 28 , 29 ], an evaluation index of drug-likeness, could effectively estimate the AMDE of chemicals. QED was calculated using the following equation: …”

A Network Pharmacology Approach to Uncover the Molecular Mechanisms of Herbal Formula Ban‐Xia‐Xie‐Xin‐Tang

“…Oral administration is the main route of administration for TCM. However, it is limited by the drug's ADME (absorption, distribution, metabolism, and elimination) characteristics [ 12 , 46 , 47 ]. The poor ADME properties are largely accountable for drug failure in exerting pharmacodynamic effect on target site in vivo [ 12 ].…”

“…This measure of drug-likeness is calculated by integrating eight physicochemical properties of molecules [ 49 ]: (1) molecular mass, (2) number of hydrogen bond donors (HBDs), (3) octanol–water partition coefficient (ALOGP), (4) number of hydrogen bond acceptors (HBAs), (5) number of rotatable bonds (ROTBs), (6) number of aromatic rings (AROMs), (7) molecular polar surface area (PSA), and (8) number of structural alerts (ALERTS). Our previous work suggested that QED values can be used for the assessment of some ADME characteristics straightforwardly [ 46 ].…”

Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches

“…Ponzoni et al [116] used both engineered and learned molecular descriptors to construct a robust machine learning model with a collected dataset comprising 202 molecules. Wang et al and Yang et al [117,118] attempted to solve the data imbalance problem by a modified RF algorithm and various sampling methods, respectively. Both studies used molecular descriptors and conducted feature selection, but interestingly, Yang et al [118] designed the workflow to find optimal feature sets and the optimal ensemble model set.…”

“…Specifically, most predictive models comprise hundreds to thousands of small chemistry datasets that cannot cover enough chemical space [76,118,188]. Moreover, the data is usually dispersed to many literatures [117,118,122,124,125,[128][129][130][134][135][136][137][138][139][140][141][142][143][144]150,155,[159][160][161][162][163][164][165][166][167][168][169]173,176,179,183], is unbalanced, and has cutoff ambiguity challenges [118,167]. Furthermore, the bioactivity assay data is strongly biased to its platform, and has an intrinsic experimental error which disrupts accurate prediction [189].…”