DNA diagnostics are important technologies in molecular and cellular biology. By allowing identification of specific sequences, DNA-based diagnostics potentially provide more accurate and rapid results than protein- or antigen-based diagnostics, primarily because phenotypic changes come much later than changes in genotype. Despite this advantage, there are fewer diagnostic or imaging systems that target DNA than those targeting proteins, antibodies, or antigens.Each type of DNA-based diagnostic has its own, unique set of limitations; however, most can be attributed to issues related to sequence restriction, signal detection, specificity, or some combination thereof. For example, while PCR-based methods allow amplification and assessment of specific DNA sequences, they lack the ability to report information of specific cells, or cell types, within the heterogeneous pool of cells typically found in a tumor biopsy. In addition, none of the currently available DNA detection methods has the potential to be utilized in living cells, a disadvantage which limits the potential applications.The work presented here describes the design and development of a new methodology for the detection of specific double-stranded DNA sequences. This detection method is based on the concept that two inactive fragments of a reporter protein, each coupled to engineered zinc finger DNA-binding motifs, are able to reassemble and form an active complex in the presence of a predefined DNA sequence. This system, designated sequence-enabled reassembly (SEER), can achieve single base-pair specificity, and has the potential to be utilized in living cells.In this dissertation, we discuss the efforts from constructing to refining the system, as well as the future applications of SEER in diagnostics and therapeutics. Chapter I will provide an introduction to DNA detection methods, on which the principles of the SEER system are based. Chapter II describes the design and construction of an enzymatic SEER system, SEER-LAC, using beta-lactamase as the enzyme. In Chapter III, we outline the in vitro characterization of the SEER-LAC system, followed by its optimization in Chapter IV. Chapter V illustrates the efforts to develop SEER system for mammalian cell culture applications. In the final chapter, the future developments and applications of SEER are discussed.

Cette thèse propose de nouveaux outils pour la caractérisation locale de familles de protéines au niveau de la séquence et de la structure. Nous introduisons les fragments en contact (CF) comme des portions de structure conciliant localité spatiale et voisinage séquentiel. Nous montrons qu'ils bénéficient d'une meilleure prédictibilité de structure depuis la séquence que des fragments contigus ou encore que des paires de fragments qui ne seraient pas en contact en structure. Pour comparer structuralement ces CF, nous introduisons l'ASD, une nouvelle mesure de similarité ne nécessitant pas d'alignement préalable, respectant l'inégalité triangulaire tout en étant tolérante aux décalages de séquences et aux indels. Nous montrons notamment que l'ASD offre des meilleures performances que les scores classiques de comparaison de fragments sur des tâches concrètes de classification non-supervisée et de fouille structurale. Enfin, grâce à des techniques d'apprentissage automatique, nous mettrons en œuvre la détection de CF à partir de la séquence pour l'identification de protéines de virus avec l'outil VIRALpro développé au cours de cette thèse. / This thesis investigates the local characterization of protein families at both structural and sequential level. We introduce contact fragments (CF) as parts of protein structure that conciliate spatial locality together with sequential neighborhood. We show that the predictability of CF from the sequence is better than that of contiguous fragments and of structurally distant pairs of fragments. In order to structurally compare CF, we introduce ASD, a novel alignment-free dissimilarity measure that respects triangular inequality while being tolerant to sequence shifts and indels. We show that ASD outperforms classical scores for fragment comparison on practical experiments such that unsupervised classification and structural mining. Ultimately, by integrating the identification of CF from the sequence into a statistical machine learning framework, we developed VIRALpro, a tool that enables the detection of sequences of viral structural proteins.