Towards the Revival of Interpretable QSAR Models

Shoombuatong, Watshara; Prathipati, Philip; Owasirikul, Wiwat; Worachartcheewan, Apilak; Simeon, Saw; Anuwongcharoen, Nuttapat; Wikberg, Jarl E. S.; Nantasenamat, Chanin

doi:10.1007/978-3-319-56850-8_1

Cited by 28 publications

(41 citation statements)

References 212 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using eighteen informative PCPs could provide faster and more cost-effective models, while model developers could gain an insight into the underlying prediction processes [58,[62][63][64]; (iii) selecting a powerful method for QSP prediction. Although, iQSP displayed a superior performance over the existing methods assessed by the rigorous crossvalidation methods, there is still room for further improvements, including increasing the size of QSPs by gathering peptide sequences from various data sources, utilizing an interpretable learning algorithm, such as scoring card method [44,53], improving the interpretation of important features responsible for the biological activity [50,64] and exploring different ML algorithms, such as extreme gradient boosting [65] or deep learning [66]. To further investigate the power of the proposed iQSP, we compared its performance with other conventional classifiers, i.e., k-nearest neighbor (k-NN), decision tree (rpart), and random forest (RF).…”

Section: Comparison With Existing Methodsmentioning

confidence: 99%

“…As noticed in Table 7, the top seven important PCPs are QIAN880137, AURR980102, ROBB760113, PRAM820101, GRAR740101, PALJ810111, and Figure 3, the superior performance of our proposed model iQSP over 10-fold CV and independent validation test might mainly be due to the following reasons: (i) Performing with multiple random sampling procedure to protect against the risk of having good predictive result by chance [39][40][41][42][43]49,50]; (ii) using an efficient feature selection method (GA-SAR) to identify m informative features from 531 PCPs. Using eighteen informative PCPs could provide faster and more cost-effective models, while model developers could gain an insight into the underlying prediction processes [58,[62][63][64]; (iii) selecting a powerful method for QSP prediction. Although, iQSP displayed a superior performance over the existing methods assessed by the rigorous cross-validation methods, there is still room for further improvements, including increasing the size of QSPs by gathering peptide sequences from various data sources, utilizing an interpretable learning algorithm, such as scoring card method [44,53], improving the interpretation of important features responsible for the biological activity [50,64] and exploring different ML algorithms, such as extreme gradient boosting [65] or deep learning [66].…”

Section: Feature Contribution Analysismentioning

confidence: 99%

See 1 more Smart Citation

iQSP: A Sequence-Based Tool for the Prediction and Analysis of Quorum Sensing Peptides Using Informative Physicochemical Properties

Charoenkwan

Schaduangrat

Nantasenamat

et al. 2019

IJMS

Self Cite

View full text Add to dashboard Cite

Understanding of quorum-sensing peptides (QSPs) in their functional mechanism plays an essential role in finding new opportunities to combat bacterial infections by designing drugs. With the avalanche of the newly available peptide sequences in the post-genomic age, it is highly desirable to develop a computational model for efficient, rapid and high-throughput QSP identification purely based on the peptide sequence information alone. Although, few methods have been developed for predicting QSPs, their prediction accuracy and interpretability still requires further improvements. Thus, in this work, we proposed an accurate sequence-based predictor (called iQSP) and a set of interpretable rules (called IR-QSP) for predicting and analyzing QSPs. In iQSP, we utilized a powerful support vector machine (SVM) cooperating with 18 informative features from physicochemical properties (PCPs). Rigorous independent validation test showed that iQSP achieved maximum accuracy and MCC of 93.00% and 0.86, respectively. Furthermore, a set of interpretable rules IR-QSP was extracted by using random forest model and the 18 informative PCPs. Finally, for the convenience of experimental scientists, the iQSP web server was established and made freely available online. It is anticipated that iQSP will become a useful tool or at least as a complementary existing method for predicting and analyzing QSPs.

show abstract

Section: Comparison With Existing Methodsmentioning

confidence: 99%

Section: Feature Contribution Analysismentioning

confidence: 99%

iQSP: A Sequence-Based Tool for the Prediction and Analysis of Quorum Sensing Peptides Using Informative Physicochemical Properties

Charoenkwan

Schaduangrat

Nantasenamat

et al. 2019

IJMS

Self Cite

View full text Add to dashboard Cite

show abstract

“…24,53 In particular, AD allows the relative estimation of the feasibility of predictions made on query compounds on the basis of how similar they are to the compounds used to train the model. There are several approaches that have been proposed to assess the AD of compounds.…”

Section: Applicability Domain Analysismentioning

confidence: 99%

Privileged substructures for anti-sickling activity via cheminformatic analysis

et al. 2018

Self Cite

View full text Add to dashboard Cite

Sickle cell disease (SCD), an autosomal recessive genetic disorder, has been recognized by the World Health Organization (WHO) as a major public health problem as it affects 300 000 individuals worldwide.Complications arising from SCD include anemia, microvascular occlusion, severe pain, stokes, renal dysfunction and infections. A lucrative therapeutic strategy is to employ anti-sickling agents that can disrupt the formation of the HbS polymer. This study therefore employed cheminformatic approaches, encompassing classification structure-activity relationship (CSAR) modeling, to deduce the privileged substructures giving rise to the anti-sickling activity of an investigated set of 115 compounds, followed by substructure analysis. Briefly, the compiled compounds were described by fingerprint descriptors and used in the construction of CSAR models via several machine learning algorithms. The modelability of the data set, as exemplified by the MODI index, was determined to be in the range of 0.70-0.84. The predictive performance was deduced by the accuracy, sensitivity, specificity and Matthews correlation coefficient, which was found to be statistically robust, whereby the former three parameters afforded values in excess of 0.7 while the latter statistical parameter provided a value greater than 0.5. An analysis of the top 20 important substructure descriptors for anti-sickling activity revealed that 10 important features were significant in the differentiation of actives from inactives, as illustrated by aromaticity/ conjugation (e.g. SubFPC287, SubFPC171 and SubFPC5), carbonyl groups (e.g. SubFPC137, SubFPC139, SubFPC49 and SubFPC135) and miscellaneous groups (e.g. SubFPC303, SubFPC302 and SubFPC275).Furthermore, an analysis of the structure-activity relationship revealed that the length of alkyl chains, choice of functional moiety and position of substitution on the benzene ring may affect the anti-sickling activity of these compounds. Thus, this knowledge is anticipated to be useful for guiding the design of robust compounds against the gelling activity of HbS, as preliminarily demonstrated in the data-driven compound design presented herein.

show abstract

“…Quantitative structure‐activity relationship models (QSAR models) apply the toolbox of chemometrics and chemoinformatics to predict the chemical, physical, or biological properties of chemicals based on a set of descriptor variables that describe the structure of the chemicals . The main areas of the application of QSAR models are the modelling of different biological activities (eg, antimalarial, antioxidant, and anti‐HIV), chemical properties (eg, chromatographic retention data and biodegradation), and toxic endpoints (eg, genotoxicity and hepatotoxicity).…”

Section: Introductionmentioning

confidence: 99%

“…Quantitative structure-activity relationship models (QSAR models) apply the toolbox of chemometrics and chemoinformatics to predict the chemical, physical, or biological properties of chemicals based on a set of descriptor variables that describe the structure of the chemicals. 1 The main areas of the application of QSAR models are the modelling of different biological activities (eg, antimalarial, 2 antioxidant, 3 and anti-HIV 4 ), chemical properties (eg, chromatographic retention data 5 and biodegradation 6 ), and toxic endpoints (eg, genotoxicity 7 and hepatotoxicity 8 ). Moreover, besides predicting the properties of certain molecular structures, the inverse task, the design of molecular structures for certain properties is achieved as well, as these QSAR models are a time-effective and economically friendly way to design new drugs.…”

Section: Introductionmentioning

confidence: 99%

Mixtures of QSAR models: Learning application domains of pKpredicto rs

Dörgő

Hamadi

Varga

et al. 2020

Journal of Chemometrics

View full text Add to dashboard Cite

Quantitative structure‐activity relationship models (QSAR models) predict the physical properties or biological effects based on physicochemical properties or molecular descriptors of chemical structures. Our work focuses on the construction of optimal linear and nonlinear weighted mixes of individual QSAR models to more accurately predict their performance. How the splitting of the application domain by a nonlinear gating network in a “mixture of experts” model structure is suitable for the determination of the optimal domain‐specific QSAR model and how the optimal QSAR model for certain chemical groups can be determined is highlighted. The input of the gating network is arbitrarily formed by the various molecular structure descriptors and/or even the prediction of the individual QSAR models. The applicability of the method is demonstrated on the pK normala values of the OASIS database (1912 chemicals) by the combination of four acidic pK normala predictions of the OECD QSAR Toolbox. According to the results, the prediction performance was enhanced by more than 15% (root‐mean‐square error [RMSE] value) compared with the predictions of the best individual QSAR model.

show abstract

Towards the Revival of Interpretable QSAR Models

Cited by 28 publications

References 212 publications

iQSP: A Sequence-Based Tool for the Prediction and Analysis of Quorum Sensing Peptides Using Informative Physicochemical Properties

iQSP: A Sequence-Based Tool for the Prediction and Analysis of Quorum Sensing Peptides Using Informative Physicochemical Properties

Privileged substructures for anti-sickling activity via cheminformatic analysis

Mixtures of QSAR models: Learning application domains of pKpredicto rs

Contact Info

Product

Resources

About