Implementation of a web based universal exchange and inference language for medicine: Sparse data, probabilities and inference in data mining of clinical data repositories

Data-mining to build a knowledge representation store for clinical decision support. Studies on curation and validation based on machine performance in multiple choice medical licensing examinations

“…For discussing the theoretical basis, we continue to use the notation standard and associated probabilistic algebra due to theoretical physicist Paul Dirac. It was developed in the 1930s to 1940s [13] for quantum mechanics (QM) [14] and author Robson and colleagues have adapted it for classical inference over several years [15][16][17][18][19][20][21][22][23]. MARPLE owes a debt to a considerable body of work going back to efforts in early data mining in bioinformatics in the 1970s, as reviewed in Refs.…”

“…MARPLE owes a debt to a considerable body of work going back to efforts in early data mining in bioinformatics in the 1970s, as reviewed in Refs. [20,23,27]. Dirac notation may be an accepted standard in physics and so its use here may please physicists, but to most readers it is still largely an unfamiliar approach and somewhat less essential for understanding here because QM is about intrinsic probabilities, and it is extrinsic probabilities that played the important role in exams (as discussed above, and in Ref.…”

“…It is also called a Semantic Triple (ST) because it has three parts, analogous to subject-verb-object (SVO) clause. The knowledge represented in these can interchange with preexisting Q-UEL applications such as POPPER [22], and DiracBuilder [23], and its sources are human experts inputted via POPPER HELPER [23] and structured data mining such as DiracMiner [23], as well as other Q-UEL applications and clinical data repositories [27,28].The knowledge from the Internet which is said to be less well curated form (Section 1.3) is captured as attribute values called XTRACTs in Q-UEL XTRACT tags generated by the automatic browser and surfer called XTRACTOR [27]. It is these XTRACTs that are the rougher knowledge elements that are extracted from source text.…”

“…This can be applied several times for each question as described below, each viewing the contextual importance of the tag from a slightly different perspective, so note that MARPLE resets back to zero the accumulative counts of N F þ , N W þ N F-, and N W-only when moving on to a new question. See Section 2.5 next, regarding v. For s ¼ 1 we have the Euler series that gives an expected information measure [20,23]…”

“…Again, if a probability value is not explicit, the value is 1. In this report, tags come from DiracMIner [23] in the structured case and XTRACTOR [27] in the unstructured case, and also from Q-UEL archives especially those that have been built up by POPPER over the past five years, which includes a lot of knowledge entered by human experts using the POPPER HELPER interface [22]. These are essentially already MARPLE KRS elements, differing only in the format for presenting probabilities, so that conversion is trivial.…”