An interview with Ali Daowd exploring the potential of literature-based discovery for cancer drug repurposing

In this interview with Ali Daowd (Dalhousie University, Halifax, Canada), we discuss how literature-based discovery be applied to understanding and treating cancers. Ali provides his insights into what this technique offers, how it can be utilized for cancer drug repurposing and what challenges need to be overcome in future studies.

Please could you introduce yourself and your organization(s) and provide a brief summary of your career to date?

I am a PhD candidate in medical informatics with the NICHE research group at Dalhousie University. I also hold a master’s degree in health informatics and a medical degree from the Royal College of Surgeons in Ireland – Bahrain (Manama, Bahrain). Given my interdisciplinary background, my research interest is to apply novel informatics-based techniques to solve complex medical problems. I have been involved in many research projects with NICHE under the supervision of Syed Sibte Raza Abidi, including developing a citizen-centered digital health platform to engage and educate the public on the prevention of chronic diseases, designing patient-facing visualizations of personal health data and, most recently, exploring semantics-enabled literature mining to automatically uncover hidden knowledge about chronic diseases and cancers.

Could you please explain how literature-based discovery is applied to understanding and treating cancers and what it offers as a research technique?

Finding and reviewing scientific publications is one of the key aspects for developing new cancer treatments, but scientists nowadays are finding it increasingly difficult to stay up-to-date with the latest scientific developments due to the rapid growth of the literature. It is reported that the doubling time of biomedical knowledge has increased from 3.5 years in 2010 to just 73 days in 2020, so there is a wealth of knowledge in the literature that goes unnoticed.

Literature-based discovery is a research paradigm that aims to automatically uncover hidden knowledge in the literature by combining text mining and semantic-based analytics. Literature-based discovery has the potential to implicitly connect and reason over segregated silos of knowledge in the literature to discover previously unknown biomedical interactions.

Literature-based discovery is premised on what is known as the ABC model, which states that if a paper describes an interaction between biomedical entities A and B, and if another independent paper identifies an interaction between entities B and C, then there is an indirect interaction between A and B that is yet to be explored. Therefore, literature-based discovery can be applied to large-scale literature to automate the discovery of new indications for existing drugs in a process known as drug repurposing.

New candidate drugs can be repurposed for cancers by mining biomedical knowledge from the literature and then applying literature-based discovery to discover indirect interactions between a drug and a cancer through their interaction with a common gene, pathway or biological process.

As a research technique, literature-based discovery can be utilized for various purposes in biomedicine, including generating new hypotheses, drug repurposing, discovering disease biomarkers and explaining the underlying molecular mechanisms causing diseases.

Recently, literature-based discovery has been utilized to develop biomedical question-answering systems by combining knowledge extracted from the literature with curated knowledge from biomedical ontologies and knowledge bases. Overall, literature-based discovery offers an innovative alternative to traditional information retrieval methods, which are not concerned with discovering implicit connections between extracted knowledge entities.

What were the main aims and take-home findings of your recent paper on utilizing literature-based discovery for cancer drug repurposing?

The main objective in this paper was to address the challenges of acquiring meaningful biomedical knowledge from the literature. Literature-based discovery relies on the output of text mining methods to identify and extract knowledge from biomedical literature. However, despite recent computational advances, most text mining methods suffer from the problem of low recall – that is, potentially useful knowledge is not being identified. Incomplete knowledge extraction from literature can hinder the discovery process in literature-based discovery.

To overcome this problem, we proposed a novel knowledge completion method by utilizing representation learning techniques to predict missing knowledge entities. Specifically, we represented biomedical knowledge extracted from the literature as a knowledge graph of nodes and semantic relations, and then employed knowledge graph embeddings to predict missing relations between nodes in the literature-based graph.

Our findings suggest that knowledge graph completion methods can enhance literature-based discovery by augmenting the incomplete knowledge extracted from biomedical literature with accurate and meaningful relation predictions. We replicated several cancer-drug repurposing discoveries published in recent studies by applying literature-based discovery on the completed knowledge graph. The same experiment was also conducted on an incomplete knowledge graph as a baseline, which failed to replicate the drug repurposing discoveries.

What limitations were there, and do you intend to do follow-up research to address this?

The main limitation of this work is the utilization of simple knowledge graph embedding models to predict missing knowledge entities in biomedical knowledge graphs. Recent advances in graph-based representation learning have resulted in developing robust deep-learning semantic models that are capable of extracting more hidden information from large-scale data.

Regarding future work, we will consider employing more sophisticated representation learning techniques to extract informative semantic features from literature-based knowledge graphs. We also plan to experiment with integrating curated biomedical knowledge to overcome the limitations of incomplete knowledge extraction from biomedical literature.

Another limitation, which is common to most literature-based discovery research, is the lack of collaboration with biomedical experts to validate the knowledge discovery outputs. Most literature-based discovery implementations remain confined within a research setting and have not seen widescale adoption among the larger scientific community. To fully realize the benefits of literature-based discovery, strong collaborations are needed between literature-based discovery developers and subject matter experts to provide more credibility to data-driven knowledge discovery approaches.

What does the future hold for utilizing AI in cancer drug repurposing?

The outbreak of the COVID-19 pandemic has accelerated the uptake of AI-based drug discovery and we now have ample evidence supporting the utilization of AI to repurpose existing drugs for new indications. AI-based drug discovery is considered a cost and time-efficient alternative to traditional de novo drug design, which is known to cost billions of dollars with very low success rates.

With advanced deep-learning models, we can now extract informative features from large-scale biomedical data more efficiently and produce reliable results in a short period of time. However, any application of AI should be considered with a healthy dose of scepticism as the ‘black box’ problem remains a significant challenge that is yet to be solved. I believe that the development of explainable AI models will be a major focus in future research to provide human-like explanations of outputs and, thereby, help to overcome the hurdles of non-explainable AI.

‍The opinions expressed in this feature are those of the interviewee/author and do not necessarily reflect the views of Future Medicine AI Hub or Future Science Group.