Section: New Results

Sequences

Random generation

In collaboration with the Simon Fraser University (Vancouver, Canada), we have explored a random generation strategy, under a Boltzmann distribution, to assess the robustness of predicted adjacencies in ancestral genomes using a parsimony-based approach. The sampling algorithm was used to estimate the Boltzmann probability of ancestral adjacencies, which was then used as a filter to weed out unsupported predictions, leading to the resolution of a large number of syntenic inconsistencies [23] .

Combinatorics of motifs

An algorithm for pvalue computation has been proposed in [40] that takes into account a Hiddden Markov Model and an implementation, SufPref , has been realized (http://server2.lpm.org.ru/bio ).

Combinatorics of clumps have been extensively studied, leading to the definition of the so-called canonic clumps. It is shown in [26] that they contain the necessary information needed to calculate, approximate, and study probabilities of occurrences and asymptotics. This motivates the development of a clump automaton. It allows for a derivation of pvalues, decreasing the space and time complexity of the generating function approach or previous weighted automata. An extension to degenerate patterns is currently realized and implemented in a collaboration with J. Holub (Praha U.) and E. Furletova (Impb ).

During her master thesis at King's College, A. Héliou and collaborators designed the first linear-time and linear-space algorithm for computing all minimal absent words based on the suffix array [6] . In a typical application, one would be interested in computing minimal absent words to compare and study genomes in linear time by considering this negative information.

In a collaboration with AlFarabi University, where M. Régnier acts as a foreign co-advisor), word statistics were used to identify mRNA targets for miRNAs involved in various cancers [7] .

Prediction and functional annotation of ortholog groups of proteins

In comparative genomics, orthologs are used to transfer annotation from genes already characterized to newly sequenced genomes. Many methods have been developed for finding orthologs in sets of genomes. However, the application of different methods on the same proteome set can lead to distinct orthology predictions.

In [38] , [14] we developed a method based on a meta-approach that is able to combine the results of several methods for orthologous group prediction. The purpose of this method is to produce better quality results by using the overlapping results obtained from several individual orthologous gene prediction procedures. Our method proceeds in two steps. The first aims to construct seeds for groups of orthologous genes; these seeds correspond to the exact overlaps between the results of all or several methods. In the second step, these seed groups are expanded by using HMM profiles.

We evaluated our method on two standard reference benchmarks, OrthoBench and Orthology Benchmark Service. Our method presents a higher level of accurately predicted groups than the individual input methods of orthologous group prediction. Moreover, our method increases the number of annotated orthologous pairs without decreasing the annotation quality compared to twelve state-of-the-art methods.