> >crunch. Engineering bacteria and jarvesting proteins in (k)g quantities
> >from fermenters is not particularly hard.
>
> Just using lots of cpu power won't solve the PFP, at least not before
> rod-logic computers are cheap.
Uh, I don't think so. From my estimations, a 50 k$ hardware investment now
will yield in a desktop device capable of traditional-MD-probing well into
the us domain (roughly a 1000x increase since 1987/88). Sans any novel
algorithms whatsoever (btw, such simple things as DMPTA have turned O(n*n)
into O(n) while maintaining accurate long-range Coulomb interactions).
Today, a hypothetical large ASIC box should reduce that amount noticeably,
and thus make ms domains accessible. Several proteins fold that quick.
This brute-force approach may be actually unnecessary. If we can guarantee
not to miss a critical carrefour in the folding process (which is not too
easy), shortcutting the brownian randomwalk in the configuration space by
a clever minimization should reduce the effort vastly. Precise endgame
potential can be extracted from empirical data.
> >> Markus Krummenacker have some well thought out plans about making an
> >> assembler from protein and dna that doesn't seem to require any software
> >
> >Doesn't _seem_. I do not see how I can engineer enzymes, and design
> >good-fit complementary surfaces without a full-blown IPFP. Btw, granted
>
> Instead of engineering new proteins, just build things by attaching
> known proteins to each other and to dna.
The idea of autoassembly is based on complementary-surface alignment and
noncovalent bonding of subblocks resulting in your desired space-filling.
Noncovalent attachements require excellent precision, even more so
covalent bonds (xref disulfide bond engineering difficulties in protein
stabilization).
ciao,
'gene