@Unpublished{GlasgowFortierConklinAllenLeherte2004,
author = {Glasgow, J. and Fortier, S. and Conklin, D. and
Allen, F. and Leherte, L.},
title = {Knowledge representation for molecular scene analysis.},
note = {},
OPTkey = {},
OPTmonth = {},
OPTyear = {2004},
OPTannote = {}
}
The proposed knowledge representations are a descriptive (for logical and numerical operations) and two depictive ones (for image operations).
(The paper reviews the basics of protein structure and crystallographic methods for studying it, which I will not summarize.)
Determination of structure is like scene analysis, drawing on LTM for structural motifs, and drawing on the rules of biochem.
Differences between the AI's database and other crystallographic
databases:
1. also has general concepts derived from analysis.
2. has links between knowledge structures through which implicit
knowledge can be derived.
3. organized for modification of knowledge structures, rather than
optimized for searching.
4. includes procedural "know-how" of experts and algorithms.[7]
There is a part-of hierarchy as well as a class, subclass and instance hierarchy in the KR. [9] This allows for easy reasoning at different levels of abstraction, as well as inheritance. This is being made into a KB using the Nial Frame Language.[14]
Quillian introduced the idea of the "intersection search" of a semantic network to see how related 2 nodes are.[9]
Frames are superior to object sets for this application in the following way: Frames allow for the dynamic creation of lew levels of parts or abstractions, where object-oriented systems do not. [14]
The three levels of computational imagery: [15]
Spatial motifs are described with description logics. One concept subsumes another if its extension is a superset of the other's. Amino acid subsumes valine because all instances of valine are instances of amino acids. Since description logics have trouble with part-of relationships, they crafted SDL, the spatial description logic. [16] It adds the concept of a symbolic image. It's concept terms in a multi-dimensional space corresponding to the has-parts and rel-coordinates slots in teh frames. Protein motifs are described in SDL.
The IMEM (Image MEMory) system uses an incremental concept formation technique. [17] It's being used to discover protein motifs.
Motifs can eb anticipated top-down if the motif is associated with the primary structure (the amino acid sequence). The bottom-up way to do it involves the visual representation.
IMEM has discovered sensible motifs. [18]
The SYSTEM: ORCRIT [19] does topographical analysis of electron density maps (EDMs) to get a skeleton of critical points.