Explainable artificial intelligence in high-throughput drug repositioning for subgroup stratifications with interventionable potential

Enabling precision medicine requires developing robust patient stratification methods as well as drugs tailored to homogeneous subgroups of patients from a heterogeneous population. Developing de novo drugs is expensive and time consuming with an ultimately low FDA approval rate. These limitations make developing new drugs for a small portion of a disease population unfeasible. Therefore, drug repositioning is an essential alternative for developing new drugs for a disease subpopulation. This shows the importance of developing data-driven approaches that find druggable homogeneous subgroups within the disease population and reposition the drugs for these subgroups. In this study, we developed an explainable AI approach for patient stratification and drug repositioning. Contrast pattern mining and network analysis were used to discover homogeneous subgroups within a disease population. For each subgroup, a biomedical network analysis was done to find the drugs that are most relevant to a given subgroup of patients. The set of candidate drugs for each subgroup was ranked using an aggregated drug score assigned to each drug. The proposed method represents a human-in-the-loop framework, where medical experts use the data-driven results to generate hypotheses and obtain insights into potential therapeutic candidates for patients who belong to a subgroup. Colorectal cancer (CRC) was used as a case study. Patients' phenotypic and genotypic data was utilized with a heterogeneous knowledge base because it gives a multi-view perspective for finding new indications for drugs outside of their original use. Our analysis of the top candidate drugs for the subgroups identified by medical experts showed that most of these drugs are cancer-related, and most of them have the potential to be a CRC regimen based on studies in the literature.

Efficient prediction of drug-drug interaction using deep learning models

A drug-drug interaction or drug synergy is extensively utilised for cancer treatment. However, prediction of drug-drug interaction is defined as an ill-posed problem, because manual testing is only implementable on small group of drugs. Predicting the drug-drug interaction score has been a popular research topic recently. Recently many machine learning models have proposed in the literature to predict the drug-drug interaction score efficiently. However, these models suffer from the over-fitting issue. Therefore, these models are not so-effective for predicting the drug-drug interaction score. In this work, an integrated convolutional mixture density recurrent neural network is proposed and implemented. The proposed model integrates convolutional neural networks, recurrent neural networks and mixture density networks. Extensive comparative analysis reveals that the proposed model significantly outperforms the competitive models.

Turning genome-wide association study findings into opportunities for drug repositioning

Drug development is a very costly and lengthy process, while repositioned or repurposed drugs could be brought into clinical practice within a shorter time-frame and at a much reduced cost. Numerous computational approaches to drug repositioning have been developed, but methods utilizing genome-wide association studies (GWASs) data are less explored.

The past decade has observed a massive growth in the amount of data from GWAS; the rich information contained in GWAS has great potential to guide drug repositioning or discovery. While multiple tools are available for finding the most relevant genes from GWAS hits, searching for top susceptibility genes is only one way to guide repositioning, which has its own limitations.

Here we provide a comprehensive review of different computational approaches that employ GWAS data to guide drug repositioning. These methods include selecting top candidate genes from GWAS as drug targets, deducing drug candidates based on drug-drug and disease-disease similarities, searching for reversed expression profiles between drugs and diseases, pathway-based methods as well as approaches based on analysis of biological networks. Each method is illustrated with examples, and their respective strengths and limitations are discussed. We also discussed several areas for future research.

Meta-path Based Prioritization of Functional Drug Actions with Multi-Level Biological Networks

Functional drug actions refer to drug-affected GO terms. They aid in the investigation of drug effects that are therapeutic or adverse. Previous studies have utilized the linkage information between drugs and functions in molecular level biological networks. Since the current knowledge of molecular level mechanisms of biological functions is still limited, such previous studies were incomplete. We expected that the multi-level biological networks would allow us to more completely investigate the functional drug actions. We constructed multi-level biological networks with genes, GO terms, and diseases. Meta-paths were utilized to extract the features of each GO term. We trained 39 SVM models to prioritize the functional drug actions of the various 39 drugs. Through the multi-level networks, more functional drug actions were utilized for the 39 models and inferred by the models. Multi-level based features improved the performance of the models, and the average AUROC value in the cross-validation was 0.86. Moreover, 60% of the candidates were true.

Prediction of drug synergy score using ensemble based differential evolution

Prediction of drug synergy score is an ill-posed problem. It plays an efficient role in the medical field for inhibiting specific cancer agents. An efficient regression-based machine learning technique has an ability to minimise the drug synergy prediction errors. Therefore, in this study, an efficient machine learning technique for drug synergy prediction technique is designed by using ensemble based differential evolution (DE) for optimising the support vector machine (SVM). Because the tuning of the attributes of SVM kernel regulates the prediction precision. The ensemble based DE employs two trial vector generation techniques and two control attributes settings. The initial generation technique has the best solution and the other is without the best solution. The proposed and existing competitive machine learning techniques are applied to drug synergy data. The extensive analysis demonstrates that the proposed technique outperforms others in terms of accuracy, root mean square error and coefficient of correlation.

Rare Diseases: Drug Discovery and Informatics Resource

A rare disease refers to any disease with very low prevalence individually. Although the impacted population is small for a single disease, more than 6000 rare diseases affect millions of people across the world. Due to the small market size, high cost and possibly low return on investment, only in recent years, the research and development of rare disease drugs have gradually risen globally, in several domains including gene therapy, enzyme replacement therapy, and drug repositioning. Due to the complex etiology and heterogeneous symptoms, there is a large gap between basic research and patient unmet needs for rare disease drug discovery. As computational biology increasingly arises researchers' awareness, the informatics database on rare disease have grown rapidly in the recent years, including drug targets, genetic variant and mutation, phenotype and ontology and patient registries. Along with the advances of informatics database and networks, new computational models will help accelerate the target identification and lead optimization process for rare disease pre-clinical drug development.

Transcriptional response networks for elucidating mechanisms of action of multitargeted agents

Drug discovery is moving away from the single target-based approach towards harnessing the potential of polypharmacological agents that modulate the activity of multiple nodes in the complex networks of deregulations underlying disease phenotypes. Computational network pharmacology methods that use systems-level drug-response phenotypes, such as those originating from genome-wide transcriptomic profiles, have proved particularly effective for elucidating the mechanisms of action of multitargeted compounds. Here, we show, via the case study of the natural product pinosylvin, how the combination of two complementary network-based methods can provide novel, unexpected mechanistic insights. This case study also illustrates that elucidating the mechanism of action of multitargeted natural products through transcriptional response-based approaches is a challenging endeavor, often requiring multiple computational-experimental iterations.