there are all kinds of interesting puzzles buried in the grail blood test cancer diagnostic paper

A reasonable solution to most of them:

There is a lot more signal in DNA methylation (epigenetic modifications) than has previously been understood.

An example:

The MSK-impact cancer tissue biopsy test (cost: $3,500) looks at 0.06% of the genome and runs it through the sequencer 800 times.

The Grail test looks at 0.6% of the genome plus methylation and runs it through the sequencer 140 times.

The tissue test should be much simpler to get signal from: ~all of the material going through the sequencer is from the cancer.

For the blood diagnostic you are presumably putting a lot of non-cancer DNA fragments into your sequencer--more noise.

So they are probably doing some fragment-size selection prep at the front end and maybe generating data from the yield in that process,

and they are likely applying better machine-learning against their larger target region which compounds with the methylation data...

But it is very surprising that with only 2x the sequencer utilization of a cancer tissue biopsy DNA test, you can detect early stage cancer with precision greater than 99% from a blood draw.

It suggests to me that there is another wave of discovery coming:

genome-coupled methylation data from next gen sequencers will provide researchers and clinicians with marvelously precise diagnostic instruments.

It also suggests that grail could bring the cancer diagnostic blood test to market on today's tech stack at $8,000

A highly specific blood-draw cancer diagnostic for $2,000 could be 3 years away.