a******n
发帖数: 100 | 1 https://venturebeat.com/2018/12/03/deepminds-alphafold-wins-casp13-protein-
folding-competition/
AlphaFold handily outguns competing solutions in terms of performance. It
beat 98 competitors in the Critical Assessment of Structure Prediction (CASP
) protein-folding competition in Cancun, where it successfuly predicted the
structure of 25 out of 43 proteins. (The second-place system could only
predict three protein structures.) More impressively, it managed to forecast
its first protein structures in a matter of hours — magnitudes faster than
previous systems.
Improving the scientific community’s understanding of protein folding could
lead to more effective diagnoses and treatment of diseases such as
Parkinson’s and Alzheimer’s, the DeepMind team noted, as these are
believed to be caused by misfolded proteins. And it could aid in protein
design, leading to protein-secreting bacteria that make wastewater
biodegradable, for instance, and enzymes that can help manage pollutants
such as plastic and oil. | a******n
发帖数: 100 | 2 有感兴趣的吗? 我是做计算的，ML,DL。可站内联系。 | x****6
发帖数: 4339 | 3 它的算法有物理解释吗？ 还是完全是一个black box?
【在 a******n 的大作中提到】
: 有感兴趣的吗? 我是做计算的，ML,DL。可站内联系。
| a******n
发帖数: 100 | 4 Haven't seen their paper yet.
In general: no, it's just black box; but those can be added as features if
it helps. I saw there are chemists on their team, so probably they added.
However, remember AlphaGo, it beats any programs (and even human itself)
that based on what human
has thought as principle / explanation (所谓"棋理")
【在 x****6 的大作中提到】
: 它的算法有物理解释吗？ 还是完全是一个black box?
| z*******x
发帖数: 103 | 5 how much does its performance come from its super-speed?
In principle, protein structure is a kind of pattern, like Go.
So, IBM machin can be super good at structure prediction, like what it does
for GO.
Disclaim: I do not know much about protein structure | a******n
发帖数: 100 | 6
you mean just hardware alone? to some extent, but not that much.
does
(then why it's not IBM who did it (AlphaGo/Fold)?)
No, if you followed what happened in Deep Learning, ANN, DNN in recent years
, you should know it's more because of the advance in ML/DL algorithms.
From your comments looks like you think they are still using the similar
technology in the era of DeepBlue.
【在 z*******x 的大作中提到】
: how much does its performance come from its super-speed?
: In principle, protein structure is a kind of pattern, like Go.
: So, IBM machin can be super good at structure prediction, like what it does
: for GO.
: Disclaim: I do not know much about protein structure
| I******i
发帖数: 203 | | B*******L
发帖数: 10 | 8 他们使用的deep learning算法并不是原创的。关于这个算法的第一篇论文在 https://
journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005324
这篇论文在2018年初拿了PLoS Comp. Biol. Research Prize in Breakthrough/
Innovation. See
https://blogs.plos.org/blog/2018/06/01/winners-of-the-2018-plos-
computational-biology-research-prize/
CASP
the
forecast
than
could
【在 a******n 的大作中提到】
: https://venturebeat.com/2018/12/03/deepminds-alphafold-wins-casp13-protein-
: folding-competition/
: AlphaFold handily outguns competing solutions in terms of performance. It
: beat 98 competitors in the Critical Assessment of Structure Prediction (CASP
: ) protein-folding competition in Cancun, where it successfuly predicted the
: structure of 25 out of 43 proteins. (The second-place system could only
: predict three protein structures.) More impressively, it managed to forecast
: its first protein structures in a matter of hours — magnitudes faster than
: previous systems.
: Improving the scientific community’s understanding of protein folding could
|
|
