contextual predicates in protein folding..
| Case number: | 699969-990188 |
| Topic: | General |
| Opened by: | Seagat2011 |
| Status: | Open |
| Type: | Suggestion |
| Opened on: | Thursday, August 18, 2011 - 00:55 |
| Last modified: | Saturday, August 20, 2011 - 22:40 |
In Adwait Ratnaparkhi's 1998 Ph.D. dissertation, "maximum entropy models for natural language ambiguity resolution", he discusses the use of contextual predicates to aid in (machine) learning. Each probabilistic sample from the solution space is presented as a function (i.e. a contextual predicate) and its informational useful is weighted by a set of Yes or No questions, as to whether the solution's "context" also exists in the solution space.
This allows users to enter highly detailed queries without the risk that important solution sets that do not exactly match the query will not be found.
?
This information I think will be very useful to recipe writers and the developers of foldit.
If you say so.
Perhaps an example might make it more clear?
Sure.
"The Stanford NLP (Natural Language Processing) Group" - http://nlp.stanford.edu/downloads/lex-parser.shtml
..Since the story engine is a probabilistic system, it searches for the best match for this query, not just for strings that meet all the criteria. For instance, a document that referenced a communication involving [Location:Woods], [Context:woods], [subtext:shy], [George:Exhausted,Max:Determined] is likely to be ranked highly, even if it does not fall in the data range specified.
Application in protein folding: Using this approach (saving not only the best rebuild, but also near-to-best rebuild solutions), an intelligent recipe writer may be able to "avoid" local minima.
And as an example of this, if my foldit rebuild tool returns a rebuild, z, with a high mini-Rosetta score, k, then the optimal solution p is the most uncertain distribution that satisfies the mini-Rosetta(z) constraints on feature expectations.
Is this some sort of SPAM ?