3 ms·

Is sequencing 'hard' computational work? Could we use it for proof-of-work?

by boppo1 5y ago

teekert 5y ago
Aligning all the short reads a (Sequencing By Synthesis) Next Generation Sequencer (to be very specific) produces (hundreds to thousands of millions) to a reference genome is hard computational work indeed, no idea if you can use it for proof-of-work but I like where this is going... But do you not need some kind of definition of correct (like hash needs many leading 0's) for PoW? How would that work for aligning reads to a reference genome? Maybe it could, find the position of the read is hard (I mean, not really for a modern CPU but relatively... I think... not an expert on that), verifying how correct it is, is not hard (I think). Maybe we should ask the BWA devs [0]. [0]: https://github.com/lh3/bwa https://github.com/lh3/bwa
- jghn 5y ago
  I’d be in line to purchase some HengCoin. It’s an interesting idea but you also raise a good point that alignment is probabilistic. It is the step that most benefits from cpu parallelization so would be cool to explore at least
  - teekert 5y ago
    What do you mean by alignment being probabilistic? If you mean that some reads align faster and you could game the system, then perhaps we should tie the system to hospital based verification of the input data. Ie, hospital publishes sequencing data (perhaps using fully homomorphic encryption to keep this very sensitive data private), someone aligns it, someone (or the hospital) verifies it, aligner gets paid coin. The verification of the alignment being easier than the alignment itself is analogous to blockchain verification being easier than actually signing a block, right?
    - jghn 5y ago
      Not quite the best phrase I'll admit. It's identifying the best fit. That's not the same as identifying truth. For instance, my recollection is that bwa-mem has (had?) some small amount of nondeterminism in it.
      - teekert 5y ago
        Ah yes, it would be easy to confirm a good alignment, identifying if any alignment is the best there is would be more difficult I guess, indeed.

Is sequencing 'hard' computational work? Could we use it for proof-of-work?

by boppo1 5y ago

Is sequencing 'hard' computational work? Could we use it for proof-of-work?

Aligning all the short reads a (Sequencing By Synthesis) Next Generation Sequencer (to be very specific) produces (hundreds to thousands of millions) to a reference genome is hard computational work indeed, no idea if you can use it for proof-of-work but I like where this is going... But do you not need some kind of definition of correct (like hash needs many leading 0's) for PoW? How would that work for aligning reads to a reference genome? Maybe it could, find the position of the read is hard (I mean, not really for a modern CPU but relatively... I think... not an expert on that), verifying how correct it is, is not hard (I think). Maybe we should ask the BWA devs [0]. [0]: https://github.com/lh3/bwa https://github.com/lh3/bwa

jghn 5y ago
I’d be in line to purchase some HengCoin. It’s an interesting idea but you also raise a good point that alignment is probabilistic. It is the step that most benefits from cpu parallelization so would be cool to explore at least
- teekert 5y ago
  What do you mean by alignment being probabilistic? If you mean that some reads align faster and you could game the system, then perhaps we should tie the system to hospital based verification of the input data. Ie, hospital publishes sequencing data (perhaps using fully homomorphic encryption to keep this very sensitive data private), someone aligns it, someone (or the hospital) verifies it, aligner gets paid coin. The verification of the alignment being easier than the alignment itself is analogous to blockchain verification being easier than actually signing a block, right?
  - jghn 5y ago
    Not quite the best phrase I'll admit. It's identifying the best fit. That's not the same as identifying truth. For instance, my recollection is that bwa-mem has (had?) some small amount of nondeterminism in it.
    - teekert 5y ago
      Ah yes, it would be easy to confirm a good alignment, identifying if any alignment is the best there is would be more difficult I guess, indeed.