One in two people in the UK will get cancer during their lifetime. Many cancers are increasingly treatable but better targeted drugs are needed to reduce rates of recurrence and side-effects. Artificial intelligence is now being used to tackle cancer by distinguishing patterns in the way that our genes and their protein products interact, in how these interactions affect the cell, and in ways they can be exploited as potential drug targets.
It is estimated that DNA damage occurs up to 10,000 times per cell per day in humans. Most damage is repaired and badly damaged cells are generally destroyed or forced to stop dividing before any problems occur.
Some genetic damage however, may result in unchecked, important changes to the cell, allowing a cell to carry on dividing even as the cell starts to experience significant damage such as structural instability. Cancers start when particular patterns of genetic damage cause cells to grow and divide out of control.
Targeting the right cells
All cancers share important features. The genetic damage affects specific cell functions that increase proliferation rate, decrease death rate and create a growth-promoting environment.
Despite the increased growth, cancer cells are fragile and targeted additional damage that would be tolerated by a normal cell may tip a cancerous cell into cell death. The trick in cancer therapy is to identify how to damage cells so that healthy ones survive and cancer cells are destroyed.
But cancer is complicated. Cells have around 20,000 genes and the profile of genetic damage in each type of cancer and each person’s cancer is distinct. Worse still, most genetic damage is caused by the cancer but has little impact on the disease. It is this mix of lots of data with occasional patterns mixed in that makes cancer an ideal area for tackling with artificial intelligence.
These artificial intelligence methods rely on pooling all the information that is known about how genes code for proteins and how these proteins interact whether in humans or yeast etc to form vast data networks.
The machine learning algorithm is trained by being given patterns of network data that are associated with cell death and then uses the model to identify similar patterns elsewhere in the network.
Killing cancerous cells while sparing the healthy cells
One example of the patterns sought after relates to the interactions between just a single pair of genes where the cell needs at least one of the genes to be functioning properly in order to survive. This phenomena is called synthetic lethality and offers a new way of treating cancer that may reduce collateral damage. If one of the genes is damaged by a cancer it may be possible to suppress the protein products of the other gene therapeutically, to kill the damaged cancer cell while sparing healthy cells.
A typical experimental approach to identifying these synthetic lethal interactions is to damage different pairs of genes in yeast cultures or flies and then observe the impact on growth. These experiments are essential but they are laborious. Furthermore even though many human genes are similar to those in lower organisms, the direct results from experiments on yeasts and flies don’t always tally very well with similar experiments on human cancer cell-lines.
AI to test more human gene pairs
At present there are only around 500 known pairs of human genes which act in this way but artificial intelligence is predicting many more that can be tested. This paper published in the Journal of Integrative Bioformatics reviews some of the ways that artificial intelligence and machine learning are being used to predict and identify synthetic lethal interactions that might be exploited as potential cancer drug targets.
Read the original article here