Predicting clinically promising therapeutic hypotheses using tensor factorization

A comparative analysis of computational drug repurposing approaches: proposing a novel tensor-matrix-tensor factorization method

Efficient drug discovery relies on drug repurposing, an important and open research field. This work presents a novel factorization method and a practical comparison of different approaches for drug repurposing. First, we propose a novel tensor-matrix-tensor (TMT) formulation as a new data array method with a gradient-based factorization procedure. Additionally, this paper examines and contrasts four computational drug repurposing approaches-factorization-based methods, machine learning methods, deep learning methods, and graph neural networks-to fulfill the second purpose. We test the strategies on two datasets and assess each approach's performance, drawbacks, problems, and benefits based on results. The results demonstrate that deep learning techniques work better than other strategies and that their results might be more reliable. Ultimately, graph neural methods need to be in an inductive manner to have a reliable prediction.

Development of a human genetics-guided priority score for 19,365 genes and 399 drug indications

Studies have shown that drug targets with human genetic support are more likely to succeed in clinical trials. Hence, a tool integrating genetic evidence to prioritize drug target genes is beneficial for drug discovery. We built a genetic priority score (GPS) by integrating eight genetic features with drug indications from the Open Targets and SIDER databases. The top 0.83%, 0.28% and 0.19% of the GPS conferred a 5.3-, 9.9- and 11.0-fold increased effect of having an indication, respectively. In addition, we observed that targets in the top 0.28% of the score were 1.7-, 3.7- and 8.8-fold more likely to advance from phase I to phases II, III and IV, respectively. Complementary to the GPS, we incorporated the direction of genetic effect and drug mechanism into a directional version of the score called the GPS with direction of effect. We applied our method to 19,365 protein-coding genes and 399 drug indications and made all results available through a web portal.

A Knowledge Graph-Enhanced Tensor Factorisation Model for Discovering Drug Targets

The drug discovery and development process is a long and expensive one, costing over 1 billion USD on average per drug and taking 10-15 years. To reduce the high levels of attrition throughout the process, there has been a growing interest in applying machine learning methodologies to various stages of drug discovery and development in the recent decade, especially at the earliest stage - identification of druggable disease genes. In this paper, we have developed a new tensor factorisation model to predict potential drug targets (genes or proteins) for treating diseases. We created a three-dimensional data tensor consisting of 1,048 gene targets, 860 diseases and 230,011 evidence attributes and clinical outcomes connecting them, using data extracted from the Open Targets and PharmaProjects databases. We enriched the data with gene target representations learned from a drug discovery-oriented knowledge graph and applied our proposed method to predict the clinical outcomes for unseen gene target and disease pairs. We designed three evaluation strategies to measure the prediction performance and benchmarked several commonly used machine learning classifiers together with Bayesian matrix and tensor factorisation methods. The result shows that incorporating knowledge graph embeddings significantly improves the prediction accuracy and that training tensor factorisation alongside a dense neural network outperforms all other baselines. In summary, our framework combines two actively studied machine learning approaches to disease target identification, namely tensor factorisation and knowledge graph representation learning, which could be a promising avenue for further exploration in data-driven drug discovery.

DDIT: An Online Predictor for Multiple Clinical Phenotypic Drug-Disease Associations

Background: Drug repurposing provides an effective method for high-speed, low-risk drug development. Clinical phenotype-based screening exceeded target-based approaches in discovering first-in-class small-molecule drugs. However, most of these approaches predict only binary phenotypic associations between drugs and diseases; the types of drug and diseases have not been well exploited. Principally, the clinical phenotypes of a known drug can be divided into indications (Is), side effects (SEs), and contraindications (CIs). Incorporating these different clinical phenotypes of drug–disease associations (DDAs) can improve the prediction accuracy of the DDAs.

Methods: We develop Drug Disease Interaction Type (DDIT), a user-friendly online predictor that supports drug repositioning by submitting known Is, SEs, and CIs for a target drug of interest. The dataset for Is, SEs, and CIs was extracted from PREDICT, SIDER, and MED-RT, respectively. To unify the names of the drugs and diseases, we mapped their names to the Unified Medical Language System (UMLS) ontology using Rest API. We then integrated multiple clinical phenotypes into a conditional restricted Boltzmann machine (RBM) enabling the identification of different phenotypes of drug–disease associations, including the prediction of as yet unknown DDAs in the input.

Results: By 10-fold cross-validation, we demonstrate that DDIT can effectively capture the latent features of the drug–disease association network and represents over 0.217 and over 0.072 improvement in AUC and AUPR, respectively, for predicting the clinical phenotypes of DDAs compared with the classic K-nearest neighbors method (KNN, including drug-based KNN and disease-based KNN), Random Forest, and XGBoost. By conducting leave-one-drug-class-out cross-validation, the AUC and AUPR of DDIT demonstrated an improvement of 0.135 in AUC and 0.075 in AUPR compared to any of the other four methods. Within the top 10 predicted indications, side effects, and contraindications, 7/10, 9/10, and 9/10 hit known drug–disease associations. Overall, DDIT is a useful tool for predicting multiple clinical phenotypic types of drug–disease associations.

Preclinical validation of therapeutic targets predicted by tensor factorization on heterogeneous graphs

Incorrect drug target identification is a major obstacle in drug discovery. Only 15% of drugs advance from Phase II to approval, with ineffective targets accounting for over 50% of these failures1-3. Advances in data fusion and computational modeling have independently progressed towards addressing this issue. Here, we capitalize on both these approaches with Rosalind, a comprehensive gene prioritization method that combines heterogeneous knowledge graph construction with relational inference via tensor factorization to accurately predict disease-gene links. Rosalind demonstrates an increase in performance of 18%-50% over five comparable state-of-the-art algorithms. On historical data, Rosalind prospectively identifies 1 in 4 therapeutic relationships eventually proven true. Beyond efficacy, Rosalind is able to accurately predict clinical trial successes (75% recall at rank 200) and distinguish likely failures (74% recall at rank 200). Lastly, Rosalind predictions were experimentally tested in a patient-derived in-vitro assay for Rheumatoid arthritis (RA), which yielded 5 promising genes, one of which is unexplored in RA.

Are drug targets with genetic support twice as likely to be approved? Revised estimates of the impact of genetic support for drug mechanisms on the probability of drug approval

Despite strong vetting for disease activity, only 10% of candidate new molecular entities in early stage clinical trials are eventually approved. Analyzing historical pipeline data, Nelson et al. 2015 (Nat. Genet.) concluded pipeline drug targets with human genetic evidence of disease association are twice as likely to lead to approved drugs. Taking advantage of recent clinical development advances and rapid growth in GWAS datasets, we extend the original work using updated data, test whether genetic evidence predicts future successes and introduce statistical models adjusting for target and indication-level properties. Our work confirms drugs with genetically supported targets were more likely to be successful in Phases II and III. When causal genes are clear (Mendelian traits and GWAS associations linked to coding variants), we find the use of human genetic evidence increases approval by greater than two-fold, and, for Mendelian associations, the positive association holds prospectively. Our findings suggest investments into genomics and genetics are likely to be beneficial to companies deploying this strategy.

The growth of human genetics resources has the potential to help us develop better drugs. By looking at whether and how historical drug approvals could have been predicted from our current knowledge of human genetics, we can validate this approach and assess which types of genetic evidence are most likely to be useful in guiding drug discovery. Validation is important because we are often uncertain about the biological mechanisms behind genetic variants linked to disease. Most associated variants do not occur within protein-coding regions of the genome, and it is difficult to tell which of many nearby genes is contributing to disease risk. In this paper, we confirm previous correlations between genetic evidence and historical drug approvals. We find genetic evidence from severe genetic disorders and from genetic variants that alter protein sequence is more strongly associated with historical approvals. We offer statistical approaches for prioritizing new drug candidates based on whether their mechanisms are supported by human genetic evidence.