Mobile Menu

Using machine learning to model the fitness of cancer cells and predict drug resistance

Recently, researchers have developed a machine learning approach that can predict the fitness trajectories of cancer cells in response to a particular treatment.

The ever-changing nature of tumours is one of the biggest challenges impacting the development of effective cancer treatments. The fitness of cancer cells is not set in stone and can be influenced by the environment. For example, cancer cells that thrive in surroundings saturated with chemotherapy drugs are likely to behave differently in an environment without those drugs. Therefore, predicting how tumours evolve in response to certain treatments is difficult.

Copy number alterations (CNAs) are somatic changes to chromosome structure that result in the gain or loss of copied DNA sections. CNAs promote tumour progression by altering gene expression levels. Fitness landscapes map the genotypes of a set of mutants onto a fitness-related phenotype, such as drug resistance, to inform about molecular and population genetic mechanisms that drive evolutionary change. Until now, the significance of CNAs on genomic fitness landscapes in cancer has not been clear, with scientists doubting their importance in cancer progression. This was mainly due to a lack of studies incorporating time-series single-cell sampling of polyclonal populations.

Machine learning predicts cancer fitness

Researchers at Memorial Sloan Kettering Cancer Centre and the University of British Columbia Cancer recently showed that a machine learning approach could be used to accurately predict how human tumours will evolve in the future. The team studied triple negative breast cancer patients and generated 42,000 genomes from time-series single-cell sequencing of breast tissue. It was found that mutations in the TP53 gene altered the tumour fitness landscape and that these mutations were reproduced over many clones from CNA-based genotypes.

It is known that somatic mutations in the TP53 gene are one of the most frequent alterations in human cancers. However, this study demonstrated that the mutated TP53 clonal fitness was defined by CNAs, providing crucial evidence for the important role that CNAs play in tumour evolution.

Furthermore, it was found that the treatment of tumours with chemotherapy led to drug-resistant cells which had distinct CNAs. Then, when chemotherapy was removed, the presence of drug-resistant cells also decreased. In fact, drug-resistant cells became outnumbered by the original drug-sensitive cells. This indicates a fitness cost of treatment resistance, which is linked to both CNAs and therapeutic resistance in tumours.

Cancer treatment innovation

There were several factors that allowed the scientists to create this novel model to predict how the cancer cells would evolve:

  • Realistic patient models, called xenografts, were used. A xenograft refers to a tissue or organ that is derived from a species that is different from the recipient of the specimen. In this study, human cancers were removed from patients and transplanted into mice. The scientists were then able to analyse these tumour models repeatedly over a three-year period.
  • Single-cell sequencing technology was applied to document the genetic makeup of thousands of individual cancer cells in the tumours at the same time.
  • A machine learning tool, called fitClone, applied population genetics to cancer cells in the tumours. This was used to describe how a population will evolve depending on the frequency of individual cell fitness levels.

Essentially, the team have developed an approach that could predict whether a patient’s tumour is likely to stop responding to a particular treatment and identify the cells that are likely to be responsible for the relapse. The prospect for similar methods is hugely exciting as it could help scientists intervene earlier, before the cancer has had a chance to evolve or develop resistance. Identifying certain clones in a tumour and predicting how they are likely to evolve will help highly tailored treatments to be delivered at optimal times and, ultimately, improve cancer patient outcomes.

Dr Shah, Chief of Computational Oncology at Memorial Sloan Kettering, explained:

“Population genetic models of evolution match up nicely to cancer, but for a number of practical reasons it’s been a challenge to apply these to the evolution of real human cancers. Also, historically, the field has focused on the cancer from a single snapshot. This study is an important conceptual advance because it demonstrates that the fitness trajectories of cancer cells are predictable and reproducible. We also show that it’s possible overcome some of the previously experienced barriers”.

Image credit: FreePik user7350813

Share this article