Enhanced QSAR model performance by integrating structural and gene expression information

Modeling study for predicting altered cellular activity induced by nanomaterials based on Dlk1-Dio3 gene expression and structural relationships

Nanomaterials have been widely applied and developed due to its unique physicochemical characteristics, such as their small size. The environmental and biological effects caused by nanomaterials have raised concerns. In particular, some nanometal oxides have obvious biological toxicity and pose a major safety problem. The prediction model established by combining the expression levels of key genes with quantitative structure-activity relationship (QSAR) studies can predict the biotoxicity of nanomaterials by relying on both structural information and gene regulation information. This model can fill the gap of missing mechanisms in QSAR studies. In this study, we exposed A549 cells and BEAS-2B cells to 21 nanometal oxides for 24 h. Cell viability was assessed by measuring absorbance values using the CCK8 assay, and the expression levels of the Dlk1-Dio3 gene cluster were measured. By using the theoretical basis of the nano-QSAR model and the improved principles of the SMILES-based descriptors to combine specific gene expression and structural factors, new models were constructed using Monte Carlo partial least squares (MC-PLS) for the biotoxicity of the nanometal oxides on two different lung cells. The overall quality of the nano-QSAR models constructed by combining specific gene expression and structural parameters for A549 and BEAS-2B cells was better than that of the models constructed based on structural parameters only. The coefficient of determination (R²) of the A549 cell model increased from 0.9044 to 0.9969, and the Root Mean Square Error (RMSE) decreased from 0.1922 to 0.0348. The R² of the BEAS-2B cell model increased from 0.9355 to 0.9705, and the RMSE decreased from 0.1206 to 0.0874. The model validation proved the proposed models have a good prediction, generalization ability and model stability. This study offers a new research perspective for the toxicity assessment of nanometal oxides, contributing to a more systematic safety evaluation of nanomaterials.

How doppelgänger effects in biomedical data confound machine learning

Machine learning (ML) models have been increasingly adopted in drug development for faster identification of potential targets. Cross-validation techniques are commonly used to evaluate these models. However, the reliability of such validation methods can be affected by the presence of data doppelgängers. Data doppelgängers occur when independently derived data are very similar to each other, causing models to perform well regardless of how they are trained (i.e., the doppelgänger effect). Despite the abundance of data doppelgängers in biomedical data and their inflationary effects, they remain uncharacterized. We show their prevalence in biomedical data, demonstrate how doppelgängers arise, and provide proof of their confounding effects. To mitigate the doppelgänger effect, we recommend identifying data doppelgängers before the training-validation split.

Prediction of Adverse Drug Reactions by Combining Biomedical Tripartite Network and Graph Representation Model

As one of the primary contributors to high clinical attrition rates of drugs, toxicity evaluation is of critical significance to new drug discovery. Unsurprisingly, a vast number of computational methods have been developed at various stages of development pipeline to evaluate potential adverse drug reactions (ADRs). Despite previous success of these methods on individual ADR or certain drug family, there are great challenges to toxicity evaluation. In this study, a novel strategy was developed to predict the drug-ADR associations by combining deep learning and the biomedical tripartite network. This heterogeneous network contains biomedical linked data of three entities, for example, drugs, targets, and ADRs. For the first time, GraRep, a deep learning method for distributed representations, is introduced to learn graph representations and identify hidden features from the tripartite network which are further used for ADR prediction. Through this approach, drug-ADR associations could possibly be discovered from a systemic perspective. The accuracy of our method is 0.95 based on internal resource validation and 0.88 based on external resource validation. Moreover, our results show the prediction accuracy using the tripartite network is better than the one with bipartite network, suggesting the model performance can be improved with further enrichment on information. According to the result of 10-fold cross validation, the deep learning model outperforms two traditional methods (topology-based measures and chemical structure-based measures). Additionally, predictive models are also constructed using other deep learning methods, and comparable results are achieved. In summary, the biomedical tripartite network-based deep learning model proposed here proves to offer a promising solution for prediction of ADRs.

An integrated quantitative structure and mechanism of action-activity relationship model of human serum albumin binding

Background

Traditional quantitative structure-activity relationship models usually neglect the molecular alterations happening in the exposed systems (the mechanism of action, MOA), that mediate between structural properties of compounds and phenotypic effects of an exposure.

Results

Here, we propose a computational strategy that integrates molecular descriptors and MOA information to better explain the mechanisms underlying biological endpoints of interest. By applying our methodology, we obtained a statistically robust and validated model to predict the binding affinity to human serum albumin. Our model is also able to provide new venues for the interpretation of the chemical-biological interactions.

Conclusion

Our observations suggest that integrated quantitative models of structural and MOA-activity relationships are promising complementary tools in the arsenal of strategies aiming at developing new safe- and useful-by-design compounds.

Electronic supplementary material

The online version of this article (10.1186/s13321-019-0359-2) contains supplementary material, which is available to authorized users.

The advancement of multidimensional QSAR for novel drug discovery - where are we headed?

INTRODUCTION

The Multidimensional quantitative structure-activity relationship (multidimensional-QSAR) method is one of the most popular computational methods employed to predict interesting biochemical properties of existing or hypothetical molecules. With continuous progress, the QSAR method has made remarkable success in various fields, such as medicinal chemistry, material science and predictive toxicology. Areas covered: In this review, the authors cover the basic elements of multidimensional -QSAR including model construction, validation and application. It includes and emphasizes the very recent developments of multidimensional -QSAR such as: HQSAR, G-QSAR, MIA-QSAR, multi-target QSAR. The advantages and disadvantages of each method are also discussed and typical examples of their application are detailed. Expert opinion: Although there are defects in multidimensional-QSAR modeling, it is still of enormous help to chemists, biologists and other researchers in various fields. In the authors' opinion, the latest more precise and feasible QSAR models should be further developed by integrating new descriptors, algorithms and other relevant computational techniques. Apart from being applied in traditional fields (e.g. lead optimization and predictive risk assessment), QSAR should be used more widely as a routine method in other emerging research fields including the modeling of nanoparticles(NPs), mixture toxicity and peptides.