Question: How do the programs analyse cancer?

Zena Hira answered on 16 Jun 2014:

Well good question.
I can give an one sided answer since I am a computer scientists and not a biologist… So what i am doing involves using machine learning to find patterns in data. Think of a simple example:
You have a number of images with animals and houses. You can very easily put them in categories.

Now, what I am doing involves a computer doing this automatically. So I am writing a program that can do that. How that works is i give it many images of animals and it fills out some parameters that these pictures have. Parameters are also called features and they can be things animals and houses have or don’t have, .i.e animals have legs, houses are big, etc. Then i give it pictures of houses and animals and it “learns” the parameters. (animals have a face, houses have doors. Houses are taller than animals. Houses are bigger. Animals have legs. Houses don’t have legs etc.)
This is called the learning stage.

Then is it is the classification stage. During the learning stage the program creates a so-called model and then it classifies new images based on that model. If I then give it a picture of a horse it should be able to distinguish that it is an animal and not a house.

So what I am doing is something similar but instead of houses and horses I only have numbers, called expressions, for a number of genes, in this case the features, the biologists got from people who have cancer and i also have a “label” for each patient. It could be dead or alive, responding or not responding to treatment, age, etc. This is the same as house and animals labels.

Then I do the same by writing a program that can statistically find genes that could help me distinguish between the labels. Given that we have a lot of genes we need to find the smaller set of genes that can help us distinguish between the labels. If we find something important we take it to the biologists and they try to find ways to improve medicine or treatments. That means that if a number of genes play more role in the correct classification of the labels, biologists need to investigate further