18 ms·
FOrward prediction of protein function from its physical structure is still considered impossibly hard if done de novo- IE, assuming no external knowledge. In
by dekhn 11y ago
FOrward prediction of protein function from its physical structure is still considered impossibly hard if done de novo- IE, assuming no external knowledge. In principle, one could simulate proteins and extract this kind of information, but it's not computationally accessible.
Folding the protein is only one part of it. This part, while not "solved", has seen an enormous amount of progress in the past few years, such that we can often predict the coarse-grained 3D structure of a protein, although not the fine details in most cases.
Understanding the enzymatic action of proteins typically requires simulating chemistry at the quantum (bonds breaking and forming) level around the active site.
Since biologists are more interested in getting results quickly, rather than solving the fundamentally hard problems (the former gets grant money more easily), and because structures that are similar tend to have similar functions, they tend to use structural homology- similarity to a protein of known function- to infer the function of a protein.
Although much of my prior work was designed to address the forward prediction problem, I have acknowledged that shortcuts produce more valuable data. And often times, that does involve finding a gene whose protein product behavior is cryptic, and then using structural homology, and other indirect methods, to refine the function of the protein.
As for the "computer science terms", I can mention that after working with large distributed systems for a long time, I treat debugging them a lot more like how biologists deal with cryptic proteins than trying to understand them from first principles. I often run "experiments" by injecting things into the distributed systemns, and monitoring them, much like scientists monitor proteins using fluorescence.