Human drug-pathway association prediction based on network consistency projection

They present a novel computational method for drug-pathway association prediction based on known drug-pathway associations. The association between a drug and a pathway needs to be examined to not only explain the cause and enable the identification, therapy, and diagnosis of a human disease. Though, biological studies and clinical trials require substantial time and resources to identify drug-pathway associations. Considerable research attention has been devoted to many scientists have developed computer models to predict the future interactions of drug–pathway organizations. They proposed a novel computing approach known as the Network Consistency Projection for Human Drug-Pathway Association (NCPHDPA). This method was based on the drug pathway target wherein biologically related drugs appear to interact with pathway targets in identical diseases and vice versa. They computed the pathway-pathway-interaction similarity of drugs sharing similarities on the basis of pairwise Jaccard similarity and then computed the drug-drug–interaction similarity of drugs sharing similar drug targets based on Jaccard similarity. The system was combined because of the cosine similarity drug network, the pathway cosine resemblance network, and the interaction network for recognized drug-pathway. NCPHDPA was a parameter less solution and did not require negative tests. Notably, NCPHDPA could be used to predict drugs without any known related pathway. Test results showed that their proposed NCPHDPA method with LOOCV achieved a high ROC of AUC = 0.7479, and with10-fold CV obtained ROC of AUC = 0.7566. The Result of ROC (AUC) comparison of NCPHDPA with other methods, such as SIMCCDA LOOCV (AUC = 0.7364), LOMDA LOOCV (AUC = 0.6729) and DMTHNDM LOOCV (AUC = 0.50.00) obtained. The robust predictive capability of the NCPHDPA was demonstrated in three case studies on drugs involved in pathways, cancer pathways, and hepatocellular carcinoma. Few attempts have been made to compared with other methods, their proposed NCPHDPA method had reliable predictive performance. The results yielded some interesting findings as that interaction of these proteins can cause a change in its associated pathway, leading to the onset of cancer.

In recent years, researchers have become increasingly interested in novel drug-pathway associations by using heterogeneous biological data. The provision of new drugs by using heterogeneous biological data is important not only to understand various drug reactions and processes for molecular interactions, but also to facilitate the production of new medicines and human disease therapies. In the field of pharmaceuticals this approach has led to the concept of one drug, one goal." This has attracted the emphasis on the role of few primary genes interacting with drugs. This relationship reveals how many drugs affect the body's pathways or associated drug with cancer pathways. The emergence of disease development from a sequence of disruptions in the human body's global pathways has been described. The discovery of drug-pathway associations deserves addition research. attention and is one of the system-based exploration challenges. Due to the time consuming, costly and intensive efforts needed to research a variety of pathways and determine whether a chemical or pathway in the cellular system communicates, computational approach can be developed to expect the possible interactions of drug pathways to understand the mode of action of medicines. Drug-pathway prediction remains in its infancy. However, multiple pathway research approaches have recently been proposed. Two solutions are applicable in modern research. The first focuses on mathematical algorithms. The other alternative avoids pharmaceutical pathway interferences; for example, an integrative factor analysis model for drug-pathway association inference (iFad) approach involves two types of data: drug sensitivity and gene expression data. Furthermore, over three types of machine learning approaches are used respectively to classify drug pathways. The descriptions of these three algorithms have been implemented as follows. Drug-pathway associations have been predicted through the use of three different learning methods, including a bipartite local model approach (BLM), Gaussian Interact Profiles kernel (GIP) method process and Graph-based Semi Supervised Learning method (GBSSL). Drugs are mainly described based on their similarities in chemical structure and functional groups and predicted due to their associated genetic expression patterns and semantic features. The problem encountered in this approach is that the sample number must be significantly lower than the typical profiles. The second approach uses enhanced data analysis based on drug genomics databases, such as PharmGKB and Drug Bank, to link disease pathways to chemicals.However, this approach inapplicable to provide the important associations, including excretion, which may be helpful in the study of drug interactions. Most drug studies have focused mainly on time-consuming and laborious approaches to research and development. Major pharmaceutical firms spend on average $1.8 billion and one decade to launch a new drug. Most researchers working in the area of cutting-edge computational models for predicting drug–target interactions, agree that and average of cost per analysis for novel anticancer drug. The efficient manufacturing of medicines leads to high drug prices. In the USA, for example, the total cost for a new cancer treatment for a diagnosis is over 100,000 dollars. Even in developed countries, many patients are unable to bear the expense of such medications, and expense may be the most frequent explanation for medication discontinuation.

ROC curve and AUC values obtained through 5-fold CV and AUC score obtained through LOOCV for NCPHDPA.

Another challenge of network-dependent drug discovery is to explain the function of pathway associations. By using computational methods to predict future drug–pathway relationship, the drug's functions mechanism of action and potential information about side effects will be understood; such an understanding has become increasingly important in recent years. Drug pathway-target identification is usually followed by traditional drug discovery. In the pathogenesis of complex diseases, however, interactions between several functionally related biomolecules in some pathways are typically complex. In addition, medicinal products typically work on relevant courses, not just on a particular objective. Drugs and pathways have been proven to have numerous biological associations. Previous research has supported the hypothesis that can suppress hedgehog pathways and rapidly regress tumors was administered to a patient with medulloblastoma. Drug-pathway associations indicate that the drug can influence the way genes are exposed in the pathways by specific genes, proteins, which contain one or more than one gene. Drug-pathway associations may provide additional physiological and functional knowledge about a chemical compound used to analysis of complex diseases. Consequently, recognizing drug pathway associations is necessary to produce drugs. Various repositories, such as the Comparative Database of Toxic Genetic Data (CTD) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) , have compiled many well-known drug-pathway associations and cancer disease information discovered through biological experiments. The number of interactions established on account of drug-pathways association is small. In addition, finding drug-pathway association by using experimental methods time-consuming and laborious. Data on drug sensitivity profiles and a high-performance transcription dataset have become easily obtainable with the development of genomics, proteomics, and metabolomics experimental technologies Such data provide valuable information on the drug-pathway association to be identified. Therefore, developing effective methods for calculating useful information to infer possible drug pathway associations is important. Trustworthy data will validate studies and minimize the use of research and financial resources. Nevertheless, the drug-pathway associations included in databases are not highly significant. High expense is a drawback of biological experiments or clinical studies. Researchers have thus developed efficient bioinformatics computational models and methods, such as drug-pathway and drug-disease models. Investigators have mainly focused on developing new mathematical models and methods for modeling probable drug-pathway interactions. Labeled and unlabeled samples are mainly used for the creation of machine learning models. Cross-validation tests indicate strong success in model prediction.

ROC curve and AUC values obtained through 10-fold CV and AUC score obtained through LOOCV for NCPHDPA.

However, given that the concept of individual drug targets limits the research and development of new drugs, researchers have been utilized a pathway-based strategy for drug discovery. In addition, different types of data, such as drug-related chemical structure, pathway-related gene expression, and known drug and pathway associations, are integrated as characteristics representing drugs and pathways in other machine learning models. Various classifiers can then be used to train prediction models through training samples. Previous research has proposed the hypothesis that (FacPad), Bayesian sparse factor model. This kind of transcriptomic profiling data is therefore of great importance in the detection of pathways to different drugs. However, for this role there are few computational resources available. Wang, Dong-Qin et al. conducted research on methods of exploration can resolve these limitations based on pathways. This method identifies drug-pathways associations with the lasso-type penalty for the regularization term, such as Integrative Penalized Matrix Decomposition (iPaD) method. A L1L2,1-iPaD calculation model has been constructed to identify drug-pathway associations based on Integrative Penalized Matrix Decomposition. The newly developed drug-pathway recognition association gains attention because of its ability to decode action mechanisms and compound objectives. proposed in this study a new method of detection for drug-pathway association prediction called: integral graph regularization of matrix factoring (IGMF). It uses graph regularization to encode geometric data and avoid potential prediction overfitting.

The ROC curve and AUC values of their proposed methods in the 10-fold CV, 5-fold CV and LOOCV Auc score of (NCPHDPA).

Moreover, the L1-norm regularization of the objective function allows the parts-based and sparse data representation. In order to achieve this purposive effect, the drug was initially represented by the functional groups and the chemical structure's similarity. They conducted all analyses using Gaussian Interaction Profiles with the RLS method was shown to be efficient. The semi-supervised methods show good output in relation to the data set with positives much lower than negative tests, to the best of their knowledge. An interesting side finding was that GBSSL method is a traditional half-monitored method using a graphic representation of the data with a node for each sample labeled and unlabeled. In particular, they are integrating drug (and target) interaction profiles into a network of binary vectors that imply the presence or absence of interaction within that network. A possible profile of this used (GIP) kernel, for prediction drug objective interactions, using a single classifier, (kernel), Regularized Least Squares, (RLS). BLM is a supervised method used to infer drug pathway associations. In this method, the local support vector machine (SVM) classifier is used for the prediction of drug-related pathways for each drug. The SVM is then also used to forecast pathway-related drugs for each pathway. A supervised learning method is suitable in this case given that newly several public databases can provide information on accurate drug–target interactions. The training set contains a number of compounds and proteins involved in known drug–target interactions.

Drug-targeted genes associated with pathway in cancer.

The conformational pathways that link the active state pathway and inactive state pathway which is lac kinase domain were described in this works, coupled with the critical dynamics sampling technique molecular dynamics simulation. In this method, they defined and liken such a pathway with those gained for a comparable cinase i.e., c-Src, the key structural determinants characterizing such essential conformational transitions. A case study was performed on the main influenza drug aligned with oseltamivir. The pathways found by this approach are close to those identified by extremely costly computational methods in previous studies. In this article, they used a cost-effective method called 'pathway-docking' to detect potential routes of the ligand-receptor entering a protein binding pocket, which is a binding energy surface.The activation pathway caused by silicon analysis of the Histaprodifen experimental method. The study modeling may provide new insights into the procedure of ligand binding at the molecular level. To demonstrate silicon research, experimental studies should be carried out in future, which may allow the observation of structural changes in the ligand binding process.

Computed AUROC score of individual drug-pathway association.

Several researchers have used random walks on the two networks to predict possible drug-disease associations and drug-drug interactions. However, the defect of the random walking model persists in this procedure. Although the aforementioned methods are currently producing reasonably good results, many have limits. Some supervised methods cannot predict drugs without known associated pathways or disease. In this research, they propose a new calculation method called the Network Consistency Projection Drug-Pathway Associations (NCPHDPA), which is based on the prediction of possible drug-pathway associations. The NCPHDPA provides excellent predictive efficiency through the inclusion of proven drug-pathway interaction networks, cosine drug networks, and drug-pathway Jaccard similarity networks. NCPHDPA does not need parameters and has a clear benefit if known drug-pathway associations are insufficient experimentally validated. They conducted analyses using leave-one-out cross-validation (LOOCV) and 10-fold cross-validation (10-fold CV), which distributes 10 parts of data sets and runs experiments. Then, they implemented 5-fold cross-validation (5-fold CV) to demonstrate NCPHDPA's prediction performance. Related studies have shown that known information from associations can produce models with high (AUC) efficiency in the prediction of drug–pathway association based on shared drug-pathway targets. They used known associations to predict novel associations and produced successful outcomes. They presented a new drug and pathology representation based on the projected consistency of networks. Then, we suggested the cosine similitude and integrated pharmaceutical similarity pathway.

Prediction of protein–protein interaction using STRING v.11.0. Arrowhead in red, shows no interaction with proteins.

Their work has five contributions: (1) They computed pathway-pathway interaction similarity (PPIs) with drugs sharing similarities based on pairwise Jaccard similarity.

(2) They computed drug-drug interaction similarity (DDIs) with drugs sharing similar drug targets on the grounds of Jaccard similarity.

(3) Theey used cosine similarity as a complementary measure of drug element and pathway similarity (PS).

(4) They determined integrated drug similarity (DS) and PS after determining cosine similarity. They calculated the integrated DS matrix and then established the integrated PS matrix.

(5) They computed the specific pathway space consistency projection score and then the specific drug space consistency projection score. Here, they present a novel drug-pathway representation based on the established network consistency for the prediction of drug-pathway associations.

Ghulam, A., Lei, X., Zhang, Y., & Wu, Z. (2022). Human drug-pathway association prediction based on network consistency projection. Computational Biology and Chemistry, 97, 107624.