Tuesday, May 31, 2005
Notational Standards
discussion à
The Roadmap for Semantic
technology adoption à
The next generation knowledge tools
(on recognizing response degeneracy, or singularity,
in the function to structure determination)
Sanjay and Ben
Ben description is precise and
should be the summary of a proposal that we offer on June 17th.
I have made five foundational principles from the text of Ben’s note. And
then I extend these principles.
1) Ontologies are used to extend the numerical feature vectors describing a biological system.
1.1) The representation of a biological system using feature vectors can be made using more than one methodology. Microarray analysis with data normalization is one methodology. The underlying assumption is that population dynamics, first developed by Rashevsky in the early 1960s, is sufficient as a model of biological function.
1.2) We proposed to vary our dependence on this assumption through the utility of metabolic pathway ontology where known science on response degeneracy is available. In the case where the biological phenomenon is “close” to a “tipping point” (where more than one metastable state can be reached with only small variation in environment or reactant concentration) control input to the model will be imposed so as to carry the model predictions to a description indicated by the ontological model
1.3) it is noted that on theoretical grounds that the use of a vectorization of representation from micro-array data is categorically similar to the vectorization of word or phrase occurrence in text understanding systems. The vectorization is sometimes all one needs to produce a high quality aggregation of concept indicators into a model of what is being referred to in the text, however at certain points certain phrases are used that have known ambiguation properties. At these points, good semantic extraction processes rely on rules that depend on ontological models.
2) These numerical feature vectors are then "crunched" using machine learning methods
2.1: The clustering of feature vectors can occur using any one of several methods, but again the process of clustering moves the representation through a series of response dependencies where the clustering process itself is subject to fine control in regions where the method can be measured (mathematically) to be instable (ie leading to more than one limiting distribution). 2.2: In these cases, the process can be influenced if definitive scientific knowledge has been encoded into an ontology.
3) The results are then rules representing patterns in biological systems, which involve ontological categories as well as reductionist components (genes, proteins)
3.1: The ontological categories exists at one level of organization, where as the components exist at a different level of organization.
3.2: These two levels of organization allows one to establish the algorithmic machinery for Q-SAR methods (qualitative structure activity relationship) analysis, and for the application of a Mill’s logic developed in the Soviet Union and extended by Kugler and myself (1997).
4) So, numerics are used as an intermediate stage in a process of figuring out descriptions of biological systems that involve combinations of biological processes/structures/functions.
5) Ontologies are used as the vocabulary for describing biological processes/structures/functions.
5.1: As such, the vocabulary assists in the communication of results within a science community.
5.2: The result
is ontology mediated collective intelligence.