Section: New Results
Sequence annotation
Participants : François Coste, Aymeric Antoine-Lorquin, Catherine Belleannée, Guillaume Collet, Gaëlle Garet, Clovis Galiez, Laurent Miclet, Jacques Nicolas, Valentin Wucher.
Automated Enzyme Classification by Formal Concept Analysis Guessing enzyme's functional activity from its sequence is a crucial task that can be approached by comparing the new sequences with those of already known enzymes labeled by a family class. This task is difficult because the activity is based on a combination of small sequence patterns and sequences greatly evolved over time. We have designed a classifier based on the identification of common subsequence blocks between known and new enzymes and the search of formal concepts built on the cross product of blocks and sequences for each class. Since new enzyme families may emerge, it is important to propose simultaneously a first classification of enzymes that cannot be assigned to a known family. Formal Concept Analysis offers a nice framework to set this task as an optimization problem on the set of concepts. The classifier has been tested with success on a particular set of enzymes present in a large variety of species, the haloacid dehalogenase (HAD) superfamily. [F. Coste, G. Garet, J. Nicolas] [28] , [10]
A bottom-up efficient algorithm learning substitutable languages from positive examples Based on Harris’s substitutability criterion, the recent definitions of classes of substitutable languages have led to interesting polynomial learnability results for expressive formal languages. These classes are also promising for practical applications: in natural language analysis, because definitions have strong linguisitic support, but also in biology for modeling protein families, as suggested in our previous study introducing the class of local substitutable languages. But turning recent theoretical advances into practice badly needs truly operable algorithms. We present here an efficient learning algorithm, motivated by intelligibility and parsing efficiency of the result, which directly reduces the positive sample into a small non redundant canonical grammar of the target substitutable language. Thanks to this new algorithm, we have been able to extend our experimentation to a complete protein dataset confirming that it is possible to learn grammars on proteins with high specificity and good sensitivity by a generalization based on local substitutability. [F. Coste, G. Garet, J. Nicolas] [29] , [10]
Logol: Expressive Pattern Matching in sequences. Application to Ribosomal Frameshift Modeling Logol consists in both a language for describing biological patterns, and an associated parser for effective pattern search in sequences (RNA, DNA or protein). The Logol language, based on an high level grammatical formalism (String Variable Grammars), allows to express flexible patterns (with mispairings and indels) composed of both sequential elements (such as motifs) and structural elements (such as repeats or pseudoknots). Its expressive power allows the design of sophisticated patterns such as the signature of "-1 programmed ribosomal frameshifting" (PRF) events in messenger RNA sequences. A PRF signature is a complex model composed of a slippery site followed by a pseudoknot located in a specific part of the sequence, which provides a good illustration of the Logol language power. [C. Belleannée, J. Nicolas, O. Sallou (GenOuest platform)] [27] [Online publication]
Identifying distant homologous viral sequences in metagenomes using protein structure information It is estimated that marine viruses daily kill about 20% of the ocean biomass. Identifying them in water samples is thus a biological issue of great importance. The metagenomic approach for virus identification is a challenging task since their sequences carry a lot of mutations and are very difficuly to identify by standard homology searches. The PEPS VAG project aims at establishing a novel methodology that uses structures of proteins as extra-information in order to annotate metagenomes without relying on sequence homology. In the context of the first experiments made on the metagenome of station 23 of the TARA Ocean Project, we used the structures of capsid proteins to infer the sequence signature of their fold, in order to find them in the metagenome. This work presents the methodology, the first experiments and the on-going improvements. [C. Galiez, F. Coste] [35]
Computational Protein Design: trying an Answer Set Programming approach to solve the problem The problem of Computational Protein Design aims at finding the best protein conformation to perform a given task. This problem can be reduced to an optimization problem, looking for the minimum of an energy function depending on the amino-acid interactions in the protein. The CPD problem may be easily modeled as an ASP program but a practical implementation able to work on real-sized instances has never been published. We have raised the main source of difficulty for current ASP solvers and ran a series of benchmarks highlighting the importance of finding a good upper bound estimation of the target minimum energy to reduce the amount of combinatorial search. Our solution clearly outperforms a direct ASP implementation without this estimation and has comparable performances with respect to SAT-based approaches. It remains less efficient than a recent approach by cost function networks, showing there still exists some place for improving the optimization component in ASP with more dynamical strategies. [J. Nicolas, H. Bazille] [34]
Searching for Optimal Orders for Discretized Distance Geometry The Molecular Distance Geometry Problem (MDGP) is the problem of finding the possible conformations of a molecule by exploiting available information on some distances between pairs of its atoms. When some assumptions are satisfied, the MDGP can be discretized, so that the search domain of the problem becomes a tree where each node corresponds to a candidate position for an atom. The search tree can be efficiently explored by using an interval Branch & Prune (BP) algorithm that can potentially enumerate all feasible conformations. In this context, the order given to the atoms of the molecule plays an important role, because it allows the discretization assumptions to be satisfied, and it also impacts the computational cost of the BP algorithm. We have proposed a new discretization order for protein backbones based on the optimization of certain criteria for a faster exploration of the discretized search domain. To this aim, we express the search for optimal orders by a set of logical constraints in ASP. Our comparison with previously proposed orders for protein backbones shows that this new discretization order makes BP perform better. [J. Nicolas, A. Muccherino (Genscale Team)] [43]
From analogical proportions in lattices to proportional analogies in formal concepts We provided an attempt at bridging formal concept analysis and the modeling of analogical proportions (i.e., statements of the form “a is to b as c is to d”). A suitable definition for analogical proportions in non distributive lattices is proposed and then applied to concept lattices. This enables us to compute what we call proportional analogies. In addition, we define the locally maximal subwords and locally minimal superwords common to a finite set of words. We also define the corresponding sets of alignments. We show that the constructed family of sets of alignments has the lattice structure. The study of analogical proportion in lattices gives hints to use this structure as a machine learning basis, aiming at inducing a generalization of the set of words. [L. Miclet] [32] , [37]