Hacker News new | more | comments | ask | show | jobs | submit login
Biotech firm Grail conducting large study for early breast cancer detection (sfchronicle.com)
30 points by akakievich 44 days ago | hide | past | web | favorite | 12 comments



Liquid biopsy is an exciting field but to call the current results “promising” seems like a stretch to me. The entire point of doing screening is catching early stage cancers. For these cancers, liquid biopsies miss far more than imaging in every study I’ve seen (even early non peer reviewed results). For this reason there is as of yet no evidence that drawing a blood sample can be a first step before imaging for the high risk populations where screening is done today. You still need to do imaging for those.

However, it seems like an interesting avenue for increasing the size of the screening population but then false positives become a serious problem within a low risk population.


This is definitely a get all of the data and see what sticks type of analysis. There should be some early evidence from a smaller sample set using cluster analysis that is suggestive that mutations can be classified and are predictive. Tons of issues to resolve. I would think that some biostatisticians should be able estimate the false positive rate from a theoretical basis.


My point is that there have already been a lot of results showing liquid biopsy misses a lot of the cancers in high risk populations. At this point theoretical estimates seem inferior as evidence than what we know so far from actually measuring.

At some point in the future we may understand more about what markers to look for in the blood and this may change but as of yet there are no "promising" results. Maybe I'm nitpicking but the concept of liquid biopsy is exciting and I hope it's something that humanity figures out how to do!


Do you have a reference I could check out? My point was this is looking at sequencing all of the cell free DNA and seeing post hoc if any of that data is predictive of some cancer developing. There are a wide number of issues taking that approach including it not being viable at all. Still it is an interesting data set. And as you state there are other biomarkers that may be more useful that still fall under the category of a liquid biopsy.. what we really mean here is something non-invasive like a blood draw.

I was interested in the FP rate theoretically just cause it would be interesting to understanding the screening watchful waiting type approach vs what we have now.


43% of Stage I cancers detected, 33% of breast cancers detected and called "the best result so far" by this one science blogger (for breast imaging these numbers are usually around 90):

https://blogs.sciencemag.org/pipeline/archives/2018/01/22/a-...


Fun fact: Grail (or, Marius at Grail) wrote their own scientific workflow engine in Go, with a focus on AWS as execution platform:

https://github.com/grailbio/reflow

Fo anyone interested, we reviewed it including a comparison to other Go based wf tools (the number is obviously growing):

http://gopherdata.io/post/more_go_based_workflow_tools_in_bi...


This is awesome. Biopsies today are incredibly invasive and painful. Anything that makes things less painful, safer, and cheaper is great -- means more ppl can use it and more ppl will get screened.


Do they also look at life factors to see if there are features that correlate with breast cancer incidence? E.g. like how smoking correlates with lung cancer.


It would seem like from a public health perspective, this is the more important thing to research.


I read this as Gmail at first.


I bet gmail could actually predict breast cancer quite well if it was let loose on it.

https://www.breastcancer.org/risk/factors

Your email knows how much you do of many of those things.


Maybe 10 years ago today it would be Facebook.

That said I always wondered if HID devices and mobile phones could be used to detect early symptoms of Parkinson’s and other neural or neuromuscular syndroms.




Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: