Michiel O. Noordewier

Using background knowledge to improve inductive learning: a case study in molecular biology

This work uses background knowledge to reexpress training data in a form more appropriate for inductive learning. The approach dramatically improves the results of decision-tree and neural network learning methods.<<ETX>>

PROXIMAL: A database system for the efficient retrieval of genetic information

We report on a query language, called PROXIMAL, for the efficient specification of complex queries over biological sequences. We present a data model for asking complex and structured questions of biological sequences, and show how the query language can be used. In addition, we describe a visual query metaphor intended for a biologist-friendly front end to the PROXIMAL query language.

Using knowledge-based neural networks to refine existing biological theories

Artificial neural networks have proven to be a useful technique for analyzing biological data and automatically producing accurate pattern recognizers. However, most applications of neural networks have not taken advantage of existing knowledge about the task at hand. This paper presents a method for using such problem-specific knowledge. The KBANN algorithm uses inference rules about the current biological problem, which need only be approximately correct, to initially configure a neural network. This network is then refined by analyzing sample examples and counter examples of the concept being learned. An application of KBANN to the prediction of E. coli transcriptional promoters demonstrates its superiority to alternative techniques, taken from both the machine-learning and molecular biology literatures. In addition, since KBANN uses a human comprehensible ``theory`` of the current problem to define the initial neural network topology, it is possible to extract a refined set of inference rules following training. A refined theory for promoter recognition is presented; the extracted rules are roughly as accurate on novel data as the trained neural network from which they came.

MEPS parameters and graph analysis for the use of recombination to construct ordered sets of overlapping clones

Homologous recombination can provide a basis for the construction of an ordered set of overlapping clones. The principle is to make two libraries, each in a vector that has a different selectable marker flanking the insert site. Recombination between the flanking markers, leading to a selectable phenotype, can only occur as the consequence of crossing over between inserts. The two libraries are crossed in a matrix, allowing the construction of an ordered set. The logic, akin to S. Benzers (1961, Genetics 47:403-415) for the arrangement of deletion and point mutations, has a graph theoretic formulation, which helps to cope with the complex and noisy data inherent in the physical mapping of genomes rich in repeated sequences. The minimum length of identity required for homologous recombination is called the MEPS (minimum efficient processing segment) and is a property of each recombination pathway. The amount and the type of sequence similarity required for two sequences to recombine is different from that implied by either the conservation of restriction sites or by most procedures of hybridization.

Case study of a knowledge-based system which plans molecular genetic experiments

A logic-based approach to planning recombinant DNA experiments is presented. Plans are constructed by attempting to match lists of constraints with existing molecules, with failure resulting in rule-directed decomposition of these constraints to create subproblems. Several AI techniques which are not traditionally associated with planning but which find application for this problem domain are used. These include an object-oriented paradigm to help reduce the size of the search space and to afford an approach to efficient machine learning and the use of simulation to solve aspects of the frame problem and other real-world difficulties idiosyncratic to molecular genetics. The resulting system is capable of solving tasks which could not be addressed by previous attempts at planning in this problem domain.<<ETX>>