AI cancer article

[gsun]

Not sure I understand this. It's not a very good article. It discusses two different ways PCa evolves and possible different treatments can be developed.

ox.ac.uk/news/2024-03-05-ai...

[Boywonder56]

ok ill be the first.....its ai .......dont understand and not a good article...two different ways....what you expect from a box with wires attached....

[gsun in reply to Boywonder56]

Meaning the article did not explain well.

[Boywonder56 in reply to gsun]

noted..

[skiingfiend]

The news article is pablum. You have to read the linked article.

In a nutshell, cancer can evolve in many different ways which makes it hard to treat. Using clustering, an algorithim that attempts to group like things together so they can be treated as a unit, based on comparing a set of features extracted from individual samples - they discovered they could construct 2 groupings that the samples could be classified as.

This is the same as grouping the population of the North Shore into two groups: West Van and North Van based on the feature - city that you live in. There is still 150k people on the north shore (not 2). All members of a grouping can be treated as similar for various purposes: marketing, set property taxes, etc.

So back to cancer, if the members of a group share similar features than maybe we can make or infer statements that apply to all of them, maybe we can find treatments that apply to all of them. This could simplify the handling of the complexity of the combinatorial # of ways that cancer can evolve whereas if we tried to treat each combination individually it would be unmanageable.

---

Just for illustration purposes (and this is not 100% in line with the article but it conveys the general notion) - over 200 mutations have been observed in PCa through genomic analysis. Not all PCa's will have the same set of mutations or in the same order (evolutionary tree) - some may have 20 mutations, some may have 50 mutations. Generally the longer you live the more mutations you will pick up. In the worst case there are 200! (factorial) different ways to sequence 200 mutations.

---

How big is 200! you say, it is very big:

788657867364790503552363213932185062295135977687173263294742533244359449963403342920304284011984623904177212138919638830257642790242637105061926624952829931113462857270763317237396988943922445621451664240254033291864131227428294853277524242407573903240321257405579568660226031904170324062351700858796178922222789623703897374720000000000000000000000000000000000000000000000000

So reducing this to 2 groupings that can be treated as similar for treatment purposes is very appealing.

---

Since there is no way to construct actual evolutionary trees or process problems of this size - they use simplification and reduction techniques to approximate the evolutionary data generated from processing tissue samples and work with that.

There's a lot more going on but I'm going to stop here.

---

This is a standard technique that has been around for decades and applies to all fields not just medicine. I first learned about it back it the early 80's in university.

This is early days and the findings have not proven to be useful yet.

[Bestdays]

It’s early days for AI. And, I think it has amazing potential in medicine.

[Boywonder56]

kinda like electric flying cars🤪

[skiingfiend in reply to Boywonder56]

There's nothing wrong with the approach it's brilliant. The big outstanding question is: Is it a valid and useful reduction -> and that is testable -> and we can determine if it provides a statistical significant advantage. So we will get an answer one way or the other.

So maybe a lot further along than electric flying cars.

[j-o-h-n in reply to Boywonder56]

I'll settle for flying nuns.....

Good Luck, Good health and Good Humor

j-o-h-n