Dr Anton Yuryev, Professional Services Director at Elsevier explores how a project utilising artificial intelligence is helping discover drugs for a rare type of cancer.
Artificial Intelligence (AI) has considerable potential in the life sciences and healthcare sectors, where it is increasingly contributing to R&D breakthroughs. With its capacity to reduce the time spent processing data, improve decision making and automate repeatable processes, AI promises to speed up vital research. Interest in the applications of AI is increasing. In a GlobalData1 survey, almost 200 life sciences companies vote AI “the most disruptive technology across the pharmaceutical industry.”
AI could be applied to significantly advance areas with huge volumes of data, including precision medicine and sequencing genomics. Using text mining and Natural Language Processing (NLP) to model diseases could lead to important insights – particularly in rare diseases where new therapies are urgently needed.
Employing AI in disease modelling
An example of AI-augmented research in action is a recent collaborative project seeking new therapies for a rare type of cancer2. The project sought to develop a disease model for diffuse intrinsic pontine glioma (DIPG), an aggressive brain cancer that typically affects children. DIPG is considered incurable, with a typical life expectancy of less than a year3 diagnosis.
Disease models aid researchers in understanding how a disease develops by identifying things like risk factors or triggers. This understanding can then be applied to determine treatment approaches. In this case, the project members wanted to develop a model to find drugs already approved by the FDA, bypassing the years-long approval processes for novel drugs, that could be repurposed to treat DIPG. The initial model was built using a network of interlinked biological relationships and entities (Biology Knowledge Graph)4 and other specialist software that analysed ‘OMICs data’ — evaluating differences in cellular molecules, including DNA and RNA — from real-world patients with DIPG.
AI-driven data analysis
After examining the OMICs data for protein activity to pinpoint the most active genes in DIPG, the researchers assessed these against a cancer hallmark model (Hanahan & Weinberg)5 to identify tumour-associated mutations. Text mining was then applied to existing scientific literature to understand how these mutations relate to mechanisms of disease.
To refine the algorithm, the Children’s Hospital of Philadelphia built a cloud-based program (Cavatica) composed of DIPG data of genetic mutations and gene expressions from over 30 patients. The AI sifted through 3,000 and 5,000 mutations per patient and another 2,000 to 3,000 additional genetic mutations. This volume of data could not be analysed by a single researcher, but with the processing capabilities of AI, two significant cancer hallmarks were identified.
Selecting drug candidates
After identifying the proteins classed as risk factors for DIPG tumour formation, the team sought to assess which drugs could repress the activity of these proteins. Initially, the team uncovered 637 drugs. The list was narrowed using NLP (driven by AI deep reading and text mining) to select those drugs that specifically prevented a mutation called TP53. The final stage involved using the model to look for existing FDA-approved drugs with the ability to inhibit the disease mechanism.
The team then ranked potential therapies, supplementing the model with pharmacokinetic, efficacy, and metabolising enzyme and transporter data. Given that the drugs would be administered to young children, toxicity was a particularly important consideration.
To date, the DIPG AI disease model has 19 different pathways covering cell type and differentiation state, disease state and biological processes. The model has discovered 212 treatments with the capacity to reverse protein activity within disease state. A quarter of these can inhibit the TP53 mutation via four different mechanisms.
The prognosis for AI
This collaborative AI project is just one demonstration of the potential AI holds in driving the next generation of precision medicine and drug repurposing. In the field of rare diseases and cancer, the unmet needs are so acute that scaling this kind of approach to other indications could hold much promise for patients.
Moreover, there is much more to discover. AI has a fundamental role in the future of precision medicine, rare disease treatment, and broader drug discovery. Yet, the industry is only scratching the surface of what AI – when used to augment human knowledge and nuance – can achieve. The next step for researchers is to develop more AI use cases, such as the project covered above. As use cases and successes are brought to wider attention, more organisations will embrace AI. This is a positive prognosis. It will advance our understanding of disease and help in delivering vital therapeutics to patients worldwide.
References
2: SNF Sinergia Project, partners include Elsevier and the Sinergia Consortium (consisting of DMG/DIPG Center at University Children’s Hospital Zurich; ETH Zurich; the Centre for Molecular Medicine Norway (NCMM)), and the Open Pediatric Brain Tumor Atlas project.
4: https://www.elsevier.com/solutions/biology-knowledge-graph?dgcid=RN_CM_Sourced_400001648
