Researchers have developed a computer model that can suggest how to optimally combine various drugs to combat cancer.
Every type of cancer is different, and different types require different types of treatment. People with liver cancer should get one type, which differs from what people with lung cancer receive.
In addition, each case of cancer can be caused by multiple mutations and must therefore be treated differently depending on the presence of one or another mutation.
In a new study, researchers developed a computer model that uses machine learning to determine how to most effectively treat people with specific types of cancer with combinations of anticancer drugs or even non-cancer drugs.
The computer model offers a completely new approach to cancer treatment, which can improve individual treatment and make it more relevant to the specific type of cancer.
“The goal is to identify various sensitivity and resistance patterns in possible treatments that can improve them even more. This applies both to finding new and more effective combinatorial therapies and to identifying which subgroups will benefit from specific combinations,” explains a researcher involved in developing the method, Krister Wennerberg, Professor, Biotech Research & innovation Centre (BRIC), University of Copenhagen.
The research has been published in Briefings in Bioinformatics.
Scanned 352 drugs for effectiveness against cancer
The background for the new study is the fact that existing treatments cannot effectively treat everyone with all types of cancer today.
Ovarian cancer is an example.
Today the primary treatments for people with ovarian cancer are good and very effective, but a few surviving cells usually develop resistance to the original treatment, which no longer works and then the cancer returns. When this happens, the doctors’ toolbox quickly becomes empty.
In the new study, the researchers developed a computer algorithm that can suggest which combinations of 352 drugs or drug candidates can target the cancer cells even though they resisted the first treatment attempt.
“The algorithm scans a wide range of anticancer drugs, drug candidates and non-cancer drugs, which may still have a potential anticancer effect because of their mechanisms of action,” says Krister Wennerberg.
Cancer can return
One challenge of cancer therapies is the fact that tumours comprise multiple cell subpopulations, each of which may be susceptible to certain drugs.
In addition to different types of cancer cells, healthy cells are also mixed in.
Focusing exclusively on the cancer cells, one can quickly imagine that some cells are susceptible to a specific treatment, but destroying these cells makes room for other types that are resistant.
Even though the cancer seems to be eliminated, it can return, and then the initial treatment no longer works because all the cells that were susceptible to that treatment have been eliminated and replaced by other growing cells.
Trained an algorithm to make treatment suggestions
The researchers made genetic expression profiles on cancer cells from four samples from women with ovarian cancer and mapped the genetic expression for all the types of cells in the tumour.
They then investigated how effective the 352 drugs were on the various cancer cells, to find drug combinations that would work well on the entire tumour. Specifically, the researchers compared the mechanisms of action of the drugs with the specific signalling pathways for which they were targeted.
This enabled the researchers to match the effectiveness of a treatment and the genetic expression of the specific cancer cell.
By training on the data, the algorithm later made predictions when the researchers presented it with new data on a new tumour. The algorithm could then identify cancer-selective drug combinations that would probably effectively treat that specific type of cancer based on the characteristics of the cancer cells in the tumour being studied.
“Our algorithm identified many drug combinations that would potentially work well on specific types of ovarian tumours. Further, the combinations did not seem particularly toxic to the patients and can therefore be considered to be tested in clinical practice,” explains Krister Wennerberg.
New trials will test the potential
The development of the algorithm is just the first step in the research. The next step will be to determine how to use the algorithm in clinical practice.
Krister Wennerberg wants to use the model to develop hypotheses about drug combinations that can then be investigated clinically.
The hope is that the model can find patterns in tumours and group them in such a way that specific combinatorial therapies can become standard for women who have cancer with certain characteristics.
The model can also help to identify characteristics that may be potential biomarkers in ovarian cancer studies.
The researchers are exploring both possibilities in a new study.
“In the new study, we will determine whether we can predict which treatments will work well based on simple examination of tumours. We will then test whether our predictions are valid in cell cultures and animal models,” says Krister Wennerberg, who adds that the researchers also plan to make similar algorithms for other types of cancer. The long-term goal is to test possible treatments clinically and thus finally validate the potential of the model.