Patterns of activity from remnants of ancient viruses in our genome may help predict which leukaemia patients will respond to chemotherapy.
In a new study, researchers from King's College London trained AI models on multiple types of clinical and biological data to understand whether they could predict which patients with acute myeloid leukaemia (AML) would respond to chemotherapy. They found that the behaviour related to ancient viral sequences in the genome was associated with a poor response to treatment, and that including this information made the models more accurate.
AML is a blood cancer that originates in the bone marrow. Despite using multiple tests, doctors cannot reliably predict whether patients will respond to chemotherapy. Many patients develop refractory AML, meaning they do not achieve full remission even though they receive treatment.
The study, published in Blood Neoplasia, looked at data from 271 AML patients who undertook standard chemotherapy. The researchers created three multilayer machine learning models, progressively adding different types of clinical and biological data. The final model also included information relating to the activity of remnants of ancient viruses, also known as endogenous retroviruses (ERVs).
ERVs comprise remnants of ancient viruses that infected our evolutionary ancestors millions of years in the past and became incorporated into the genome. They are found within the non-coding parts of our genome, some of which remain poorly understood and are sometimes referred to as the 'dark genome'. ERVs can influence a variety of biological mechanisms, including which genes are switched on, immune signalling, and more.
When ERV behaviour data was combined with the other data, the model's ability to predict who would respond to chemotherapy improved. This model also had fewer 'false positives' compared with other versions - that is, predicting a patient would have refractory AML when they may have responded to treatment.
The researchers believe ERV activity could be linked to 'viral mimicry', where the cell responds as though viral activity is taking place. This can trigger immune and inflammatory signalling, which may help sustain the treatment-resistant state of some AML cells. However, further research is needed to determine whether this process directly causes chemotherapy resistance.
Most cancer studies focus on conventional genes. However, the so-called dark genome - which makes up large parts of our DNA - remains overlooked. We identified a signature from ancient viral sequences within the dark genome that added valuable information to conventional gene-expression approaches for predicting chemotherapy response in acute myeloid leukaemia. Although further clinical validation is required, our findings demonstrate the potential of the dark genome to help identify high-risk patients earlier and guide more personalised treatment strategies."
Lead author Dr Mohammad Mahdi Karimi, Senior Lecturer in Bioinformatics, King's College London
Patient data was provided by the Beat AML clinical trial - a landmark precision medicine clinical trial launched in 2016 by Blood Cancer United.
