The field of genomics has been revolutionised by the breadth of data available and the development of artificial intelligence (AI) techniques that are powerful enough to process this information.

AI has been a key focus at the Festival of Genomics & Biodata since its launch in 2016. Over the years, our scientists have presented on the latest techniques in machine learning, which is a type of AI that finds and learns from patterns in statistics and data.

Dr. Anguraj Sadanandam leads the Systems and Precision Cancer Medicine team at the ICR and applies his multidisciplinary experience both in the wet-lab and computational biology to identify and test personalised therapies for different cancer types. He says, “Machine learning may be considered a new buzzword to some, but it’s been applied by researchers for a very long time. I have been working in this space over the last 15 years.”

Dr. Sadanandam will speak about "Precision Oncology Through Data Orchestration, Artificial Intelligence and Clinical/Preclinical Therapeutics" on 29 January as part of the AI stream of the Festival of Genomics & Biodata.

Beyond genomics

Previously, genomic data was the bottleneck and researchers would have to work with what was available. Now, there is a wealth of genomic data, which covers all genes and their variants. Researchers can also combine this with other -omics data ranging from proteomics, which examines data from proteins, to radiomics, which is based on data from medical radiology images, as well as clinical income information from patients.

Going forward Dr. Sadanandam says context is key. Researchers and clinicians need to be able to integrate various pieces of information to develop personalised cancer treatments. In the future, he said he hopes that when a patient comes in with a tumour, a doctor could look at it and tell them what the progression of that tumour could be 10 years from now.

PhenMap

Dr. Sadanandam’s lab has harnessed AI for cancer treatment in a number ways including PhenMap, a new tool for personalised cancer medicine.

PhenMap uses machine learning to identify cancer subtypes and biomarkers based. Dr. Sadanandam likens PhenMap’s approach to asking a question from the clinical and biological perspectives in parallel. PhenMap, short for phenotype mapping, starts by looking at patient prognosis and treatment responses, known as phenotypes. At the same time, PhenMap looks at biological data – this could be genomics or other data. Next, it integrates both sets of data to create integrated groupings of patients. These new categories are more distinct and could potentially identify patients who might benefit from certain treatments as well as new targets for drugs.

In a recent study, Dr. Sadanandam’s team showed PhenMap could identify clinically-relevant subtypes and biomarkers in breast cancer. These subtypes were associated with specific drug responses to an inhibitor currently in development. Researchers used data from mRNA in breast cancer cell lines and patient samples, though PhenMap could be applied to different types of data and other cancers. The tool also works with single cell samples rather than bulk tumour sequencing making it more accessible and suitable to the clinic.

Higher precision

AI tools like PhenMap help targeted cancer treatment become more personalised. Scientists have been able to group people with cancer into subtypes, but machine learning can pick up more complicated patterns and narrow these groups.

PhenMap groups patients in two ways: into discrete subgroups and as individuals along a spectrum. While a doctor may say a patient is in stage 1 or stage 3, AI tools could help pinpoint patients on a scale of 1 to 100, for example.

The greatest challenge in developing the next generation of personalised cancer treatments for patients is not more data or more powerful algorithms.

Dr. Sadanandam says, “We are now in a situation with an overwhelming amount of methods and techniques that can find a solution, but the real hurdle is making sure you’re getting the right information and then can take it to the clinic. As fast as artificial intelligence is evolving, the treatment protocols are not evolving.”

There are findings from other AI tools like PhenMap that could potentially be used in the clinic. However, it takes time for these findings and related technologies to gain regulatory approval and become licensed for patient use. Similarly, AI tools may identify new biomarkers, but validating them in the lab and developing clinical trials can take years.