Insights from our talk at Biomarkers 2023: The use of human-centric AI in biomarker identification in Oncology and Immunology
by Maria Tella
Biomarkers 2023 was held in Manchester, UK on the 27th and 28th of February. The event brought together a diverse group of experts, across pharma, biotech, and academic institutions.
Among the speakers was Causaly’s Co-Founder & CEO, Yiannis Kiachopoulos, who discussed ‘The use of human-centric AI in biomarker identification in Oncology and Immunology’. In his thought-provoking session, he explained how AI designed with the scientist at the heart of the technology is best suited for supporting the complex process of discovering biomarkers and can vastly accelerate time-to-market.
Outlined below are some key takeaways from Yiannis’s presentation.
AI: The next catalyst for drug discovery & development
It is well known that drug development is a lengthy and expensive process: clinical trial failures are all too frequent, and as a result, patients in the real world are left waiting for more effective treatments.
Although the timeframe can vary significantly depending on therapy area, trial design and enrollment, based on a 2021 report, it takes on average 10.5 years for a drug to progress from Phase 1 to regulatory approval.¹ For several years, biomarkers have been used in target validation to support early decision-making and infer early proof-of-concept for a compound. An effective biomarker strategy can greatly accelerate the drug development process, reducing overall time to approval by ~1.4 years.²
Alongside being used for target validation, biomarkers have many other uses in research and clinical development, from defining clinical trial inclusion/exclusion criteria, to being used as surrogate endpoints. These biomarkers, whether they are applied as diagnostic, predictive, prognostic biomarkers, all have the potential to accelerate time to approval. Now that the importance of biomarkers has been established, it begs the questions:
Why is it still so challenging for scientists to discover and validate biomarkers?
How can AI be used to support this process and shorten the timeframe even further?
Challenges in biomarker discovery: a knowledge problem
Key challenges include:
- Multidimensional biology: Biological pathways are highly complex, with a multitude of factors involved. Not only is it difficult to understand the link between a gene/protein and a pathway, but it is even more challenging considering the inter-dependencies involved.
- Hidden evidence: While a host of biomarkers are discovered each year, not many are transferred into the clinic. This is because validation is rigorous and requires demonstrated evidence on specificity, sensitivity, and viability. A way of identifying supportive evidence early on would drive efficiency in this process.
- Data overload: With the explosion of available data from a wide range of sources, researchers are not able to read, retain and continuously remain up to date on scientific developments, including research that directly or indirectly applies to their field of work.
- Bias: It is impossible for researchers to read all developments, resulting in inherent selection bias, and a focus on familiar pathways or methodologies. A tool that reads all evidence would greatly reduce this risk and allow for novel approaches to be explored.
- Poor traceability: Within a research team, it is too easy for only the successes to be recorded, with substandard records of the failures. Time and efforts are wasted when a researcher finds out that the biomarker they are investigating was previously explored and not viable. This lack of collective memory can greatly hinder fast innovation.
How can we use AI to support biomarker discovery?
Causaly Cloud is a platform that reads like a human and digests the entire corpus of scientific and clinical documents to build a biomedical knowledge graph. Our human-centered interface provides the tools for any scientist to find and qualify evidence as per their research needs.
In this case, let’s put ourselves in the shoes of a Translational Scientist, working in a team to design a clinical trial investigating a treatment for luminal A breast cancer. One milestone is to identify diagnostic biomarkers to define inclusion/exclusion criteria.
1. Exploring known and emerging diagnostic biomarkers for luminal A breast cancer
A conventional tool such as PubMed requires hours spent reading journals and extracting information on biomarkers. Causaly Cloud parses all evidence in a fraction of the time and allows Translational Scientists to quickly get a full map of all relevant biomarkers. Advanced visualizations such as the Timeline view allows exploration of the new, emerging biomarkers.
As Causaly Cloud’s AI reads all available biomedical papers, a scientist can explore the entire realm of biomarkers relevant to the disease, including those lesser studied, but may well be the biomarker to take further into research and validation.
2. Narrowing down and qualifying the biomarkers
Causaly Cloud allows scientists to narrow down and select different biomarkers based on criteria relevant to their research objectives, using simple and advanced filtering options. As well as conventional searches, such as whether a biomarker has been studied in humans, a scientist might want to leverage Causaly’s advanced reading and processing to select those with:
- A large number of publications supporting its role in luminal A breast cancer
- Strong language backing the evidence point
- Mentions in the results section vs the introduction of the paper, indicating that it is the major topic.
This allows the Translational Scientist to further bring to the surface biomarkers with high potential. Causaly Cloud is transparent and will always provide the scientist with the evidence points behind a given visualization to give scientists the autonomy to decide for themselves whether the evidence is relevant.
3. Prioritizing biomarkers unique to luminal A breast cancer
Trying to find unique biomarkers is not feasible in PubMed, as there is no search term that provides the answer; scientists must read as many papers as they can to pick out which biomarkers are specific to the luminal A subtype only and deprioritize others.
Causaly Cloud extracts all biological relationships from the literature into a disease-agnostic knowledge graph showing directional cause-and-effect. Scientists can delve into this powerful base of biomedical relationships at the click of a button to conduct a comparative analysis. A powerful visual output of unique biomarkers for luminal A breast cancer can be generated to allow deprioritization of non-specific biomarkers and those relevant for the luminal B subtype only.
With Causaly Cloud, a Translational Scientist can very quickly identify and propose diagnostic biomarkers for a luminal A breast cancer trial within a matter of minutes; in contrast, PubMed searches would have resulted in ~6,500 articles across the various steps. If a scientist had opted to read all abstracts, which is unlikely, it would take a minimum of 27 days – and this does not include processing the information. On top of time saved, they have also avoided selection bias and surfaced the unknown biomarkers that would have otherwise remained hidden. They can also share their findings with their colleagues via Causaly’s integrated workspace solution to expedite daily workflows and ensure traceability.
Through the steps discussed, Translational Scientists can leverage a human-centric AI system like Causaly to save significant time in the process of biomarker discovery. As a tool built for research teams, whether it is the Discovery, Translational Science or Preclinical Safety team, Causaly Cloud can provide the necessary tools to find and qualify evidence, and increase the success rates of preclinical programs.
Request a demo to learn more about how Causaly Cloud can be used to augment scientists’ capabilities.
References:
- BIO, 2021
- Krishna et al., 2008. AAPS J. 2008 Jun;10(2):401-9. doi: 10.1208/s12248-008-9041-8. Epub 2008 Aug 7.