Improved Machine Learning Models for Predicting Selective Compounds

Correlation analysis of modern analytical data - a chemometric dissection of spectral and chromatographic variables

The Standard Practices for Infrared Multivariate Quantitative Analysis (ASTM E1655) provide a guide for determining physicochemical properties of materials using multivariate calibration techniques applied to chemical sources that have high multicollinearity and correlated information. Partial least squares (PLS) is the most widely used multivariate regression method due to its excellent prediction capabilities and easy optimization. Initially applied to chromatographic data, PLS has also shown great results in near-infrared (NIR) and mid-infrared (MIR) spectroscopies. However, complex chemical matrices with low correlation may not be efficiently modeled using PLS or other multivariate analyses limited by grouping similar information (such as latent variables or principal components). Therefore, this study aims to evaluate the multicollinearity of different analytical techniques, such as high-temperature gas chromatography (HTGC), NIR, MIR, hydrogen nuclear magnetic resonance (1H NMR), carbon-13 nuclear magnetic resonance (13C NMR), and Fourier transform ion cyclotron resonance mass spectrometry coupled to the electrospray source in positive and negative ionization modes (ESI(±)FT-ICR). Descriptive statistics (coefficient of determination, R2) and principal component analysis (PCA) were used to identify the distribution of correlated information. Results showed that NIR and MIR spectroscopies exhibited a higher percentage of correlated variables, while 13C NMR and ESI(±)FT-ICR MS had more discrete profiles. Therefore, PLS development may be more effectively applied to NIR, MIR, and 1H NMR data, while 13C NMR and mass spectra may require other algorithms or variable selection methods in combination with PLS.

Machine-Learning-Based Predictions of Polymer and Postconsumer Recycled Polymer Properties: A Comprehensive Review

There has been a tremendous increase in demand for virgin and postconsumer recycled (PCR) polymers due to their wide range of chemical and physical characteristics. Despite the numerous potential benefits of using a data-driven approach to polymer design, major hurdles exist in the development of polymer informatics due to the complicated hierarchical polymer structures. In this review, a brief introduction on virgin polymer structure, PCR polymers, compatibilization of polymers to be recycled, and their characterization using sensor array technologies as well as factors affecting the polymer properties are provided. Machine-learning (ML) algorithms are gaining attention as cost-effective scalable solutions to exploit the physical and chemical structures of polymers. The basic steps for applying ML in polymer science such as fingerprinting, algorithms, open-source databases, representations, and polymer design are detailed in this review. Further, a state-of-the-art review of the prediction of various polymer material properties using ML is reviewed. Finally, we discuss open-ended research questions on ML application to PCR polymers as well as potential challenges in the prediction of their properties using artificial intelligence for more efficient and targeted PCR polymer discovery and development.

Development of a proteochemometric-based support vector machine model for predicting bioactive molecules of tubulin receptors

Microtubules are receiving enormous interest in drug discovery due to the important roles they play in cellular functions. Targeting tubulin polymerization presents an excellent opportunity for the development of anti-tubulin drugs. Drug resistance and high toxicity of currently used tubulin-binding agents have necessitated the pursuit of novel drug candidates with increased therapeutic potency. The design of novel drug candidates can be achieved using efficient computational techniques to support existing efforts. Proteochemometric (PCM) modeling is a computational technique that can be employed to elucidate the bioactivity relations between related targets and multiple ligands. We have developed a PCM-based Support Vector Machine (SVM) approach for predicting the bioactivity between tubulin receptors and small, drug-like molecules. The bioactivity datasets used for training the SVM algorithm were obtained from the Binding DB database. The SVM-based PCM model yielded a good overall predictive performance with an area under the curve (AUC) of 87%, Matthews correlation coefficient (MCC) of 72%, overall accuracy of 93%, and a classification error of 7%. The algorithm allows the prediction of the likelihood of new interactions based on confidence scores between the query datasets, comprising ligands in SMILES format and protein sequences of tubulin targets. The algorithm has been implemented as a web server known as TubPred, accessible via http://35.167.90.225:5000/ .

Multi-task learning models for predicting active compounds

The computational drug discovery methods can find potential drug-target interactions more efficiently and have been widely studied over past few decades. Such methods explore the relationship between the structural properties of compounds and their biological activity with the assumption that similar compounds tend to share similar biological targets and vice versa. However, traditional Quantitative Structure - Activity Relationship (QSAR) methods often do not have desired accuracy due to insufficient data of compound activity. In this paper, we focus on building Multi-Task Learning (MTL)-based QSAR models by considering multiple similar biological targets together and make shared information transfer across from one task to another, thereby improving not only the learning efficiency, but also the prediction accuracy. This paper selects 6 assay groups with similar biological targets from PubChem and builds their QSAR models with MTL simultaneously. According to the experiment results, our MTL-based QSAR models have better performance over traditional prominent machine learning algorithms and the improvements are even more obvious when other baseline models have low accuracy. The superiority of our models is also proved by Student's t-test with level of significance 5%. Moreover, this paper also explores three different assumptions on the underlying pattern in the dataset and finds that the joint feature MTL models further improve the performance of the QSAR models and are more suitable for building QSAR models for multiple similar biological targets.

Profiling Prediction of Kinase Inhibitors: Toward the Virtual Assay

Kinome-wide screening would have the advantage of providing structure-activity relationships against hundreds of targets simultaneously. Here, we report the generation of ligand-based activity prediction models for over 280 kinases by employing Machine Learning methods on an extensive data set of proprietary bioactivity data combined with open data. High quality (AUC > 0.7) was achieved for ∼200 kinases by (1) combining open with proprietary data, (2) choosing Random Forest over alternative tested Machine Learning methods, and (3) balancing the training data sets. Tests on left-out and external data indicate a high value for virtual screening projects. Importantly, the derived models are evenly distributed across the kinome tree, allowing reliable profiling prediction for all kinase branches. The prediction quality was further improved by employing experimental bioactivity fingerprints of a small kinase subset. Overall, the generated models can support various hit identification tasks, including virtual screening, compound repurposing, and the detection of potential off-targets.

An Algorithm to Identify Target-Selective Ligands - A Case Study of 5-HT7/5-HT1A Receptor Selectivity

A computational procedure to search for selective ligands for structurally related protein targets was developed and verified for serotonergic 5-HT7/5-HT1A receptor ligands. Starting from a set of compounds with annotated activity at both targets (grouped into four classes according to their activity: selective toward each target, not-selective and not-selective but active) and with an additional set of decoys (prepared using DUD methodology), the SVM (Support Vector Machines) models were constructed using a selective subset as positive examples and four remaining classes as negative training examples. Based on these four component models, the consensus classifier was then constructed using a data fusion approach. The combination of two approaches of data representation (molecular fingerprints vs. structural interaction fingerprints), different training set sizes and selection of the best SVM component models for consensus model generation, were evaluated to determine the optimal settings for the developed algorithm. The results showed that consensus models with molecular fingerprints, a larger training set and the selection of component models based on MCC maximization provided the best predictive performance.

Exhaustive proteome mining for functional MHC-I ligands

We present the development and application of a new machine-learning approach to exhaustively and reliably identify major histocompatibility complex class I (MHC-I) ligands among all 20(8) octapeptides and in genome-derived proteomes of Mus musculus , influenza A H3N8, and vesicular stomatitis virus (VSV). Focusing on murine H-2K(b), we identified potent octapeptides exhibiting direct MHC-I binding and stabilization on the surface of TAP-deficient RMA-S cells. Computationally identified VSV-derived peptides induced CD8(+) T-cell proliferation after VSV-infection of mice. The study demonstrates that high-level machine-learning models provide a unique access to rationally designed peptides and a promising approach toward "reverse vaccinology".

A novel approach to active compounds identification based on support vector regression model and mean impact value

Traditionally, active compounds were discovered from natural product extracts by bioassay-guided fractionation, which was with high cost and low efficiency. A well-trained support vector regression model based on mean impact value was used to identify lead active compounds on inhibiting the proliferation of the HeLa cells in curcuminoids from Curcuma longa L. Eight constituents possessing the high absolute mean impact value were identified to have significant cytotoxicity, and the cytotoxic effect of these constituents was partly confirmed by subsequent MTT (3-(4, 5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) assays and previous reports. In the dosage range of 0.2-211.2, 0.1-140.2, 0.2-149.9 μm, 50% inhibiting concentrations (IC50 ) of curcumin, demethoxycurcumin, and bisdemethoxycurcumin were 26.99 ± 1.11, 19.90 ± 1.22, and 35.51 ± 7.29 μm, respectively. It was demonstrated that our method could successfully identify lead active compounds in curcuminoids from Curcuma longa L. prior to bioassay-guided separation. The use of a support vector regression model combined with mean impact value analysis could provide an efficient and economical approach for drug discovery from natural products.