The extraction of medical knowledge from data mining many patient records and from authoritative natural language text on the Internet is important for clinical decision support. Here is discussed how such knowledge expressed in our Q-UEL language as semantic triples analogous to subject-verb-object, and now more elaborate semantic multiples, can respond to examination-style medical questions in natural language. Importantly, it is also discussed how our inference methods can be used to assign a probability to each answer in the set of candidate answers. We had intended to “show off” the power of our Q-UEL language by drawing on its many algorithmic and knowledge archive resources. However, in view of the simplicity of the approach used here, originally only intended to set prior probabilities, it is interesting that it is often alone sufficient to give the right answer, giving performance comparable to that of a typical medical student. Nonetheless, it still needs a large knowledge representation store, and Q-UEL has several tools to extract and integrate knowledge from (a) structured data, (b) unstructured data, and (c) by dialogue with human experts.

Numeric descriptions ('bio-informatic descriptions') of amino acid residues have been developed which will be of value whenever the quality and quantity of information in very large (i.e. 'human genome style') gene and protein sequences is to be compared or manipulated. These codes are as natural as possible by our criteria (the same principles could be used in revision of the criteria). In particular, in storing and searching large amounts of sequence data, natural codes--which relate to the properties of amino acids--can be combined with existing fast-search algorithms but introduce several advantages. The code can be assigned such that sub-selection of bits leads to compressed databases with residues defined less specifically, by classes of properties. The most compressed representation leads to the specification of a residue as polar or non-polar, while the most extended representation used at present also allows specification of, for example, glyco-asparagine and phosphoserine. Preliminary studies on both a supercomputer and smaller machines suggest a 'worst-case' speeding of approximately 4.5-fold. For more intelligent searching, coding extensions mixed with the basic sequence data give the sequence data some of the character of a computer program.