We are interested in using Inductive Logic Programming (ILP) to infer grammars representing sets of biological sequences. We call these biological grammars. ILP systems are well suited to this task in the sense that biological grammars have been represented as logic programs using the Definite Clause Grammar or the String Variable Grammar formalisms. However, the speed at which ILP systems can generate biological grammars has been shown to be a bottleneck. This paper presents a novel refinement operator implementation, specialised to infer biological grammars with ILP techniques. This implementation is shown to significantly speed-up inference times compared to the use of the classical refinement operator: time gains larger than 5-fold were observed in 4/5 of the experiments, and the maximum observed gain is over 300-fold.

We are interested in using Inductive Logic Programming (ILP) to infer grammars representing sets of protein sequences. ILP takes as input both examples and background knowledge predicates. This work is a first step in optimising the choice of background knowledge predicates for predicting the function of proteins. We propose methods to obtain different sets of background knowledge. We then study the impact of these sets on inference results through a hard protein function inference task: the prediction of the coupling preference of GPCR proteins. All but one of the proposed sets of background knowledge are statistically shown to have positive impacts on the predictive power of inferred rules, either directly or through interactions with other sets. In addition, this work provides further confirmation, after the work of Muggleton et al., 2001 that ILP can help to predict protein functions.

Grammars have been shown to be a very useful way to model biological sequences families. As both the quantity of biological sequences and the complexity of the biological grammars increase, generic and efficient methods for parsing are needed. We consider two parsers for context-free grammars: depth-first top-down parser and chart parser; we analyse and compare them, both theoretically and empirically, with respect to biological data. The theoretical comparison is based on a common feature of biological grammars: the gap – a gap is an element of the grammars designed to match any subsequence of the parsed string. The empirical comparison is based on grammars and sequences used by the bioinformatics community. Our conclusions are that: (1) the chart parsing algorithm is significantly faster than the depth-first top-down algorithm, (2) designing special treatments in the algorithms for managing gaps is useful, and (3) the way the grammar encodes gaps has to be carefully chosen, when using parsers not optimised for managing gaps, to prevent important increases in running times.

Grammatical inference consists in learning formal grammars for unknown languages when given sequential learning data. Classically this data is raw: Strings that belong to the language and eventually strings that do not. In this paper, we present a generic setting allowing to express domain and typing background knowledge. Algorithmic solutions are provided to introduce this additional information efficiently in the classical state-merging automata learning framework. Improvement induced by the use of this background knowledge is shown on both artificial and real data.

We consider inference of automata from given data. A classical problem is to find the smallest compatible automaton, i.e. the smallest automaton accepting all examples and rejecting all counter-examples. We study unambiguous automata (UFA) inference, an intermediate framework between the hard nondeterministic automata (NFA) inference and the well known deterministic automata (DFA) inference. The search space for UFA inference is described and original theoretical results on both the DFA and the UFA inference search space are given. An algorithm for UFA inference is proposed and experimental results on a benchmark with both deterministic and nondeterministic targets are provided showing that UFA inference outperforms DFA inference.

A classical problem in inferring automata from data is to find the smallest compatible automaton, i.e. the smallest automaton accepting all examples and rejecting all counter-examples. We are interested in Non-deterministic Finite Automata (NFA) inference which is better suited to Occam's razor optimization principle than Deterministic Finite Automata (DFA). We revisit the existing results on the search space in the state-merging framework, focusing on nondeterminism. We introduce unambiguous automata (UFA) inference, an intermediate between the difficult NFA inference framework and the weaker DFA inference framework. The UFA search space, though containing nondeterministic automata, shares many properties with the DFA search space. Our conclusion is that the search spaces of UFA and DFA are lattices when using adequate operators.

Nous considerons un probleme d'inference grammaticale. Celui-ci peut etre formule comme la tâche de decouvrir un concept caracterisant un ensemble de sequences, appele aussi langage, a partir de sequences exemple. Afin de caracteriser un langage, nous considerons une representation sous forme d'automates finis non deterministes (NFA). Ceux-ci, bien que plus difficilement manipulables que les automates deterministes (DFA) habituellement utilises, possedent pour avantage d'etre compactes et explicites (i. E. Interpretables par un expert du domaine d'application). Nous decrivons l'espace de recherche pour les NFAs, ainsi que pour differentes sous-classes des NFAs. On propose un algorithme pour l'inference d'automates non ambigus (UFA) qui necessite moins de donnees que les algorithmes usuels d'inference de DFA pour converger. Nous proposons aussi des methodes permettant d'introduire, dans l'inference, des connaissances provenant de l'application (pour nous la bioinformatique).

Nous proposons une nouvelle approche permettant d'apprendre des automates non deterministes pour la caracterisation et la modelisation de sequences proteiques. Cette approche est basee sur la fusion de fragments significativement similaires pour la caracterisation de la famille. Trois heuristiques d'ordonnancement des paires de fragment sont introduites, dont une utilisant la presence de contre-exemples, et un processus de generalisation, base sur l'identification des proprietes physico-chimiques des acides amines, est propose. Les premieres experimentations menees sur la caracterisation de proteines de la famille MIP montrent la pertinence des automates appris, attestee par un bon pouvoir de prediction. Mots cles : Inference d'automates, decouverte de motifs, sequences proteiques, proprietes physico-chimiques des acides amines.

This work addresses the problem of the inference of non deterministic automata (NFA) from given positive and negative samples. We propose here to consider this problem as a particular case of the inference of unambiguous finite state classifier. We are then able to present an efficient incompatibility NFA detection framework for state merging inference process.