Predictive Computational Models Research Articles

Rationalization of the activity Profile of Pyruvate Kinase Isozyme M2 (PKM2) Inhibitors using 3D QSAR.

Pyruvate kinase isozyme M2 (PKM2) was observed to be overexpressed and play a key role in cell growth and cancer cells' metabolism. During the past years, phytochemicals have been developed as new treatment options for chemoprevention and cancer therapy. Natural resources, like shikonin (naphthoquinone) and its derivatives, have emerged to be high potential therapeutics in cancer treatment. Our study aimed to design novel anti-tumour agents (PKM2 inhibitors) focusing on the shikonin scaffold with a better activity using computational methods. We applied a three-dimensional quantitative structure-activity relationship (3D-QSAR) approach using Field-based QSAR. The Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA) were performed on a series of forty shikonin derivatives, including shikonin, to develop robust models and rationalize the PKM2 inhibitory activity profile by building a correlation between structural features and activity. These predictive computational models will further help the design and synthesis of potent PKM2 inhibitors and their fast biological assessment at a low cost.

A Review of Computational Intelligence Models for Brain Tumour Classification and Prediction

This review aims to systematically analyze ML models from four aspects: type of ML technique, estimation accuracy, model comparison, and estimation context. A systematic literature review of empirical studies was conducted on the ML models published in the last decades. Fifty-one primary studies relevant to the objective of this research were revealed. After investigating these studies, five ML techniques have been employed in brain tumor classification and prediction. Ultimately, the estimation accuracy of these ML models could be regarded and accepted and outperformed non-ML models. ML models have been revealed to be useful in brain tumor classification and prediction. Genetic algorithm among the ML models achieved an accuracy of 100%. Nevertheless, ML models are still restricted in the industry, so initiative and encouragement are needed to make ML models easier. Further work is required on these ML models to verify the accuracy and consider other performance metrics other than the accuracy.

Prediction and validation of avascular tumor growth pattern in different metabolic conditions using in silico and in vitro models.

Objectives: In recent years, scientists have taken many efforts for in vitro and in silico modeling of cancerous tumors. In fact, three-dimensional (3D) cultures of multicellular tumor spheroids (MCTSs) are good validators for computational results. The goal of this study is to simulate the 3D early growth of avascular tumors using MCTSs and to compare the in vitro models with the results and predictions of a specific computational modeling framework. Using these two types of models, the importance of metabolic condition on tumor growth behavior and necrosis could be predicted. Materials and methods: We took advantage of a previously developed computational model of tumor growth (constructed by integrating a generic metabolic network model of cancer cells with a multiscale agent-based framework). Among the computational predictions is the importance of glucose accessibility on tumor growth behavior. To study the effect of glucose concentration experimentally, MCTSs were grown in high and low glucose culture media. After that, tumor growth pattern was analyzed by MTT assay, cell counting and propidium iodide (PI) staining. Results: We obviously observed that the rate of necrosis increases and the rate of tumor growth and cell activity decreases as the glucose availability reduces, which is in line with the computational model prediction.

Prediction of Protein-protein Interactions in Arabidopsis thaliana Using Partial Training Samples in a Machine Learning Framework

Background: Protein-protein interactions (PPI) play a vital role in a wide range of biological processes starting from cell-cell interactions to developmental control in all organisms. However, experimental identification of PPI is often laborious, time-consuming and costly compared to computational prediction. There are several computational prediction models in the literature based on complete training samples, but none of them dealt with the partial training samples. Objective: The objective of this work was to develop an effective PPI prediction model for Arabidopsis Thaliana using partial training samples in a machine learning framework. Methods: We proposed an effective computational PPI prediction model by combining random forest (RF) classifier and autocorrelation (AC) sequence encoding features with 1:2 ratio of positive- PPI and unknown-PPI samples. Results: We observed that the proposed prediction model produces the highest average performance scores of sensitivity (94.62%), AUC (0.92) and pAUC (0.189) with the training datasets and sensitivity (88.14%), AUC (0.89) and pAUC (0.176) with the test datasets of 5-fold crossvalidation compared to other candidate predictors based on LDA, LOGI, ADA, NB, KNN & SVM classifiers. It also computed the highest performance scores of TPR (91.82%) and pAUC (0.174) at FPR= 20% with AUC (0.948) compared to other candidate predictors. Conclusion: Overall performance of the developed model revealed that our proposed predictor might be useful to elucidate the biological function of unseen PPIs from a large number of candidate proteins in Arabidopsis thaliana.

Visual Analytics of Tuberculosis Detection Rat Performance.

The diagnosis of tuberculosis (TB) disease remains a global challenge, and the need for innovative diagnostic approaches is inevitable. Trained African giant pouched rats are the scent TB detection technology for operational research. The adoption of this technology is beneficial to countries with a high TB burden due to its cost-effectiveness and speed than microscopy. However, rats with some factors perform better. Thus, more insights on factors that may affect performance is important to increase rats' TB detection performance. This paper intends to provide understanding on the factors that influence rats TB detection performance using visual analytics approach. Visual analytics provide insight of data through the combination of computational predictive models and interactive visualizations. Three algorithms such as Decision tree, Random Forest and Naive Bayes were used to predict the factors that influence rats TB detection performance. Hence, our study found that age is the most significant factor, and rats of ages between 3.1 to 6 years portrayed potentiality. The algorithms were validated using the same test data to check their prediction accuracy. The accuracy check showed that the random forest outperforms with an accuracy of 78.82% than the two. However, their accuracies difference is small. The study findings may help rats TB trainers, researchers in rats TB and Information systems, and decision makers to improve detection performance. This study recommends further research that incorporates gender factors and a large sample size.

Numerical model for bending fatigue life estimation of carburized spur gears with consideration of the adjacent tooth effect

In this paper, a computational model for bending fatigue failure prediction of case hardened spur gears is proposed. The model consists of finite element gear and fatigue failure models, and it accounts for the actual stress–strain state at the tooth root by accounting for local elastic–plastic correction. By employing the proposed model, the effect of the adjacent tooth on bending fatigue lives is investigated. It is observed that bending fatigue lives can vary by approximately 22% when accounting for the adjacent tooth effect. Therefore, this effect should be considered when estimating bending fatigue lives of spur gears.

MDAPlatform: A Component-based Platform for Constructing and Assessing miRNA-disease Association Prediction Methods

Background: Increasing evidence has indicated that miRNA-disease association prediction plays a critical role in the study of clinical drugs. Researchers have proposed many computational models for miRNA-disease prediction. However, there is no unified platform to compare and analyze the pros and cons or share the code and data of these models. Objective: In this study, we develop an easy-to-use platform (MDAPlatform) to construct and assess miRNA-disease association prediction method. Methods: MDAPlatform integrates the relevant data of miRNA, disease and miRNA-disease associations that are used in previous miRNA-disease association prediction studies. Based on the componentized model, it develops different components of previous computational methods. Results: Users can conduct cross validation experiments and compare their methods with other methods, and the visualized comparison results are also provided. Conclusion: Based on the componentized model, MDAPlatform provides easy-to-operate interfaces to construct the miRNA-disease association method, which is beneficial to develop new miRNA-disease association prediction methods in the future.

Interpretable deep recommender system model for prediction of kinase inhibitor efficacy across cancer cell lines

Computational models for drug sensitivity prediction have the potential to significantly improve personalized cancer medicine. Drug sensitivity assays, combined with profiling of cancer cell lines and drugs become increasingly available for training such models. Multiple methods were proposed for predicting drug sensitivity from cancer cell line features, some in a multi-task fashion. So far, no such model leveraged drug inhibition profiles. Importantly, multi-task models require a tailored approach to model interpretability. In this work, we develop DEERS, a neural network recommender system for kinase inhibitor sensitivity prediction. The model utilizes molecular features of the cancer cell lines and kinase inhibition profiles of the drugs. DEERS incorporates two autoencoders to project cell line and drug features into 10-dimensional hidden representations and a feed-forward neural network to combine them into response prediction. We propose a novel interpretability approach, which in addition to the set of modeled features considers also the genes and processes outside of this set. Our approach outperforms simpler matrix factorization models, achieving R = 0.82 correlation between true and predicted response for the unseen cell lines. The interpretability analysis identifies 67 biological processes that drive the cell line sensitivity to particular compounds. Detailed case studies are shown for PHA-793887, XMD14-99 and Dabrafenib.

Artificial intelligence and machine learning methods in predicting anti-cancer drug combination effects.

Drug combinations have exhibited promising therapeutic effects in treating cancer patients with less toxicity and adverse side effects. However, it is infeasible to experimentally screen the enormous search space of all possible drug combinations. Therefore, developing computational models to efficiently and accurately identify potential anti-cancer synergistic drug combinations has attracted a lot of attention from the scientific community. Hypothesis-driven explicit mathematical methods or network pharmacology models have been popular in the last decade and have been comprehensively reviewed in previous surveys. With the surge of artificial intelligence and greater availability of large-scale datasets, machine learning especially deep learning methods are gaining popularity in the field of computational models for anti-cancer drug synergy prediction. Machine learning-based methods can be derived without strong assumptions about underlying mechanisms and have achieved state-of-the-art prediction performances, promoting much greater growth of the field. Here, we present a structured overview of available large-scale databases and machine learning especially deep learning methods in computational predictive models for anti-cancer drug synergy prediction. We provide a unified framework for machine learning models and detail existing model architectures as well as their contributions and limitations, shedding light into the future design of computational models. Besides, unbiased experiments are conducted to provide in-depth comparisons between reviewed papers in terms of their prediction performance.

Characterizing microRNA-mediated modulation of gene expression noise and its effect on synthetic gene circuits.

MicroRNAs (miRNAs) have been shown to modulate gene expression noise, but less is known about how miRNAs with different properties may regulate noise differently. Here, we investigate the role of competing RNAs and the composition of miRNA response elements (MREs) in modulating noise. We find that weak competing RNAs could introduce lower noise than strong competing RNAs. In comparison with a single MRE, both repetitive and composite MREs can reduce the noise at low expression, but repetitive MREs can elevate the noise remarkably at high expression. We further observed the behavior of a synthetic cell-type classifier with miRNAs as inputs and find that miRNAs and MREs that could introduce higher noise tend to enhance cell state transition. These results provide a systematic and quantitative understanding of the function of miRNAs in controlling gene expression noise and the utilization of miRNAs to modulate the behavior of synthetic gene circuits.

Protein structure prediction by AlphaFold2: are attention and symmetries all you need?

The functions of most proteins result from their 3D structures, but determining their structures experimentally remains a challenge, despite steady advances in crystallography, NMR and single-particle cryoEM. Computationally predicting the structure of a protein from its primary sequence has long been a grand challenge in bioinformatics, intimately connected with understanding protein chemistry and dynamics. Recent advances in deep learning, combined with the availability of genomic data for inferring co-evolutionary patterns, provide a new approach to protein structure prediction that is complementary to longstanding physics-based approaches. The outstanding performance of AlphaFold2 in the recent Critical Assessment of protein Structure Prediction (CASP14) experiment demonstrates the remarkable power of deep learning in structure prediction. In this perspective, we focus on the key features of AlphaFold2, including its use of (i) attention mechanisms and Transformers to capture long-range dependencies, (ii) symmetry principles to facilitate reasoning over protein structures in three dimensions and (iii) end-to-end differentiability as a unifying framework for learning from protein data. The rules of protein folding are ultimately encoded in the physical principles that underpin it; to conclude, the implications of having a powerful computational model for structure prediction that does not explicitly rely on those principles are discussed.

Intelligent and robust computational prediction model for DNA N4-methylcytosine sites via natural language processing

DNA N4-methylcytosine (4 ​mC) is an essential epigenetic modification and performs crucial roles in restriction-modification systems. The 4 ​mC involves many essential cellular processes, namely: correcting DNA replication and controlling DNA replication errors in the prokaryotic organism. In order to understand their biological functional mechanisms, the prediction of 4 ​mC modification is indispensable. Although computationally, it was targeted but the desired outcomes were not obtained. Thus, the development of an intelligent computational prediction system that truly expresses 4 ​mC modification sites is imperative. An efficient and high throughput discriminative intelligent computational system called “iDNA-4mC-DL” is introduced using the natural language processing method “word2vec” along with a convolution neural network. The obtained outcomes authenticated that the proposed iDNA-4mC-DL system performs outstandingly on six publicly available benchmark and independent datasets compared to current tools. It is, thus, highly estimated that the proposed model might be a more supportive and applied tool for rudimentary research and academia.

Improved Tolerance Limits by Combining Analytical and Experimental Data: An Information Integration Methodology

We propose a coherent methodology for integrating different sources of information on a response variable of interest, in order to accurately predict percentiles of its distribution. Under the assumption that one of the sources is more reliable than the other(s), the approach combines factors formed from the data into an additive linear regression model. Quantile regression, designed for quantifying the goodness of fit precisely at a desired quantile, is used as the optimality criterion in model-fitting. Asymptotic confidence interval construction methods for the percentiles are adopted to compute statistical tolerance limits for the response. The approach is demonstrated on a materials science case study that pools together information on failure load from physical tests and computer model predictions. A small simulation study assesses the precision of the inferences. The methodology gives plausible percentile estimates. Resulting tolerance limits are close to nominal coverage probability levels.

Combining high resolution mass spectrometry with a halogen extraction code to characterize and identify brominated disinfection byproducts formed during ozonation

Ozonation is widely used during water treatment but can generate a variety of toxic disinfection byproducts, especially in the presence of bromide. In the present study, our halogen extraction code was extended and modified to identify bromine isotopic patterns and combined with the R package MFAssignR in selectively identifying brominated disinfection byproducts (Br-DBPs) from high resolution mass spectra. In total, 127 Br-DBPs formed from a Suwannee River natural organic matter (SRNOM) solution were successfully detected from tens of thousands of mass spectrometry peaks. Kendrick mass defect analysis and structural characterization identified 17 structures, 15 of which were identified as brominated carboxylic acids and firstly reported here. Computational model predictions indicated that these brominated carboxylic acids may possess high toxic potencies and raise valid concerns. The adapted halogen extraction code described in this study is a powerful tool for a wider application of analyzing Br-DBPs in complex water matrices and provides an effective technique to characterize and identify these compounds in future studies.

Abstract 168: Computational model for prediction of actionable drug combinations in cancer

Abstract Synthetic essentiality presents an effective approach towards finding connected genetic targets in cancer by identifying genes that produce an inviable phenotype upon joint inactivation. However, it is less studied how synthetic lethal gene pairs may be contextualized with respect to deregulation of different cell functions and how these pairs act in different cell types and tissues. To study cancer-specific synthetic essentiality, we devised an integrated network-based algorithm to analyze relationships between tumor gene expression and patient survival in The Cancer Genome Atlas (TCGA) and expression data from the Genotype-Tissue Expression (GTEx) project. In particular, we identify gene pairs in each cancer cohort that exhibit upregulation in tumors but are associated with improved survival for patients whose tumors feature reduced expression of the gene pair. Our findings reveal variability in synthetic lethal pair activity across different cancer types, and the identified pairs connect distinct signaling and metabolic cellular pathways to different cancer types. Altogether, our findings present a pan-cancer reference for gene expression-based synthetic lethal gene pairs as well as a nuanced perspective on changes in synthetic essentiality across different cellular and tissue contexts, pointing to a new avenue for drug combination discovery oriented on targeting key pathways based on the cancer type. Citation Format: Sairahul Pentaparthi, Brandon Burgman, S. Stephen Yi. Computational model for prediction of actionable drug combinations in cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 168.

The role of mound functions and local environment in the diversity of termite mound structures

Mound structures that soil termites build have diverse morphologies. Previous observational studies documented that mounds are built to provide regulated environments for the termites that live within them and their structures are formed in ways to support this purpose under the influence of the mounds’ immediate environment. The objective of this study is to provide a methodology and a predictive computational model to investigate the reason behind the different but systematic shapes of termite mounds, considering all the relevant forces imposed on them and their thermoregulatory and gas-exchange functions. The gas-exchange function accounts for the capacity of the mound to diffuse metabolic gases generated in the mound’s underground nest, while the thermoregulatory function satisfies the connection between the underground nest and deep ground temperatures. The proposed predictive model is based on the principles of heat transfer and thermodynamics and allows optimized mechanically stable structures to freely emerge. The results indicate that, while the model is free to generate any mechanically stable structure, under the relevant environmental and metabolic conditions, it produces structures with forms and geometrical characteristics similar to those of natural mounds.Investigation of the connection between the local environment and the mound shapes indicated that the Sun and wind play an important role in the mound structural form. Mounds exposed to stronger solar irradiance exhibit cone-shaped structures that are pointed towards the Sun, while shaded mounds are observed to be vertical domes. The local wind is observed to affect the external shape of the mound by preventing them to grow tall while controlling the features of the internal structure. By investigating the similarities between structures in different regions (i.e., India, Namibia, and Brazil), it is revealed that, unlike mounds with a strong need for gas-exchange, mounds with a significant demand for thermoregulation exhibit deeper nests, thicker external walls, and well-defined cone- (as opposed to the dome-) shaped structures. Overall, the form of termite mounds is strongly correlated to both regulatory functions and local environments, and the resulting mound shape arises as a combination of these factors.

A Statistical and Computational Predictive Model of the Direct Fuel Injection System in Diesel Engines

In recent years, the high emission standards have grown the development of different strategies focused on the reduction of pollutants produced by combustion processes in energy transfer systems. For that reason, different studies have been developed to minimize fuel consumption and elevate internal combustion performance under different operating modes. Internal combustion engines are widely studied currently by means of advanced theories of thermal and fluid mechanics sciences with the aim to improve the energy transfer processes needed to transform the chemical energy generated in work during the complex fuel combustion process into the combustion chamber. Inexpensive methods have been developed to improve the internal combustion engine performance based on the understanding of chemical reactions and physical processes of mass and energy transfer. Mathematical and experimental models are employed to approximate the real working conditions, the physical phenomenon of the fuel flow injected into the combustion chamber of the internal combustion engine. Therefore, this paper proposes a predictive model that relates the fuel injection system with the combustion process and the heat transfer into the walls of the combustion chamber. External forces are considered during the internal combustion engine operation under real working conditions taking into account the dependent variables of the partial differential equations system that describes the internal combustion engine performance. A good agreement was reached between the experimental and predictive approaches. The results showed an error rate of less than 3 percent, considering a multiple linear regression model adjusted to the characterized internal combustion engine.

Geosmin and 2-methylisoborneol adsorption using different carbon materials: Isotherm, kinetic, multiple linear regression, and deep neural network modeling using a real drinking water source

High concentrations of geosmin (GSM) and 2-methylisoborneol (MIB) caused by cyanobacterial blooms are problematic issues in drinking water treatment plants on the Nak-Dong River in South Korea. The goal of this study was to identify the best-performing carbon materials to treat GSM and MIB in river water and to predict the final concentrations of GSM and MIB according to different water quality parameters using computational predictive modeling, multiple linear regression (MLR), and a deep neural network (DNN) for practical applications. Three types of powdered activated carbon (PAC) and three types of mesoporous carbon (MC) were compared in terms of their adsorption capacities using a source of drinking water from the Nak-Dong River. The highest maximum adsorption capacities were achieved when using C-PAC (1485.06 μg/g) in both distilled water and river water. C-PAC possesses a high micropore volume (0.45 cm3/g), low mesopore volume (0.11 cm3/g), and small, narrow pore size distribution, yielding optimal adsorption conditions. PACs yielded better results than MCs because dominant mesopore structures can hinder the adsorption of small molecules (0.6–0.8 nm) and allow them to pass through pores. GSM was removed more effectively than MIB because GSM is more hydrophobic and has a flatter structure. The high dissolved organic carbon concentration in the river water caused no reduction in GSM and MIB adsorption, but actually enhanced adsorption because a small portion of natural organic matter can compete with adsorption sites and provide additional adsorption sites based on the shrunk pore effect. The MLR and DNN models were used to predict the removal efficiency of C-PAC for GSM and MIB using 72 individual datasets with cross-validation for robust prediction and sensitivity analysis. MLR predicted the relationships between the input variables and GSM and MIB concentrations with an acceptable mean absolute error (MAE) of 6.43 ng/L for GSM and 5.80 ng/L for MIB. However, the DNN provided better agreement with a higher R2 value (>0.99) and lower MAE of 1.67 ng/L for GSM and 1.24 ng/L for MIB.

A novel miRNA-disease association prediction model using dual random walk with restart and space projection federated method.

A large number of studies have shown that the variation and disorder of miRNAs are important causes of diseases. The recognition of disease-related miRNAs has become an important topic in the field of biological research. However, the identification of disease-related miRNAs by biological experiments is expensive and time consuming. Thus, computational prediction models that predict disease-related miRNAs must be developed. A novel network projection-based dual random walk with restart (NPRWR) was used to predict potential disease-related miRNAs. The NPRWR model aims to estimate and accurately predict miRNA–disease associations by using dual random walk with restart and network projection technology, respectively. The leave-one-out cross validation (LOOCV) was adopted to evaluate the prediction performance of NPRWR. The results show that the area under the receiver operating characteristic curve(AUC) of NPRWR was 0.9029, which is superior to that of other advanced miRNA–disease associated prediction methods. In addition, lung and kidney neoplasms were selected to present a case study. Among the first 50 miRNAs predicted, 50 and 49 miRNAs have been proven by in databases or relevant literature. Moreover, NPRWR can be used to predict isolated diseases and new miRNAs. LOOCV and the case study achieved good prediction results. Thus, NPRWR will become an effective and accurate disease–miRNA association prediction model.

Computational Intelligence-Based Model for Mortality Rate Prediction in COVID-19 Patients.

The COVID-19 outbreak is currently one of the biggest challenges facing countries around the world. Millions of people have lost their lives due to COVID-19. Therefore, the accurate early detection and identification of severe COVID-19 cases can reduce the mortality rate and the likelihood of further complications. Machine Learning (ML) and Deep Learning (DL) models have been shown to be effective in the detection and diagnosis of several diseases, including COVID-19. This study used ML algorithms, such as Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbor (KNN) and DL model (containing six layers with ReLU and output layer with sigmoid activation), to predict the mortality rate in COVID-19 cases. Models were trained using confirmed COVID-19 patients from 146 countries. Comparative analysis was performed among ML and DL models using a reduced feature set. The best results were achieved using the proposed DL model, with an accuracy of 0.97. Experimental results reveal the significance of the proposed model over the baseline study in the literature with the reduced feature set.