Nicolas Scalzitti

Gadolinium‐based contrast agents (GBCAs) are indispensable tools in magnetic resonance imaging (MRI), yet their clinical use is limited by non‐specific tissue accumulation, low molecular specificity, and safety concerns. Protein and peptide scaffolds provide a promising alternative because they can bind metal ions with high selectivity and enable precise molecular targeting. However, identifying short peptide motifs with optimal gadolinium (Gd ³⁺ ) coordination and high relaxivity remains a major challenge. Here, we used a machine‐learning‐driven peptide evolution platform, the Protein Optimization Engineering Tool (POET), to design and optimize short Gd‐binding motifs that enhance longitudinal relaxivity ( r ₁ ). Two algorithmic strategies were tested: motif‐based and regular‐expression‐based representations. Both algorithms were trained on an initial set of 74 twelve‐amino‐acid peptides derived from natural EF‐hand scaffolds. Through two rounds of directed evolution and experimental screening, POET predicted peptides with up to a 24% increase in r ₁ ratio compared with the best natural EF‐hand. Further analysis revealed that peptides exhibiting higher relaxivity generally possessed a more negative net charge and lower isoelectric point than the buffer pH, indicating stronger electrostatic stabilization of Gd ³⁺ . Sequence enrichment analysis showed that acidic and small polar residues, particularly aspartic acid, glycine, and threonine, were selectively favored during evolution, while bulky hydrophobic and basic residues were depleted. These compositional trends align with improved solubility and enhanced metal coordination. Together, these results demonstrate a generalizable framework that integrates computational evolution with biophysical screening to discover new biologically derived Gd‐binding motifs. This approach provides a scalable route to engineer responsive, tunable, and biocompatible MRI contrast tags for precision imaging and molecular diagnostics.

Les projets de séquençage à haut débit produisent une énorme quantité de données biologiques brutes. Cependant, elles sont difficilement exploitables si elles ne sont pas annotées. Pour traiter ces données, des programmes d’annotation de génomes ont été développés, mais ces derniers sont encore trop sujet aux erreurs de prédiction, faisant de l’annotation des génomes un des défis majeurs en bio-informatique. Dans ce contexte, mes travaux de thèse s’organisent autour d’un trinôme : 1) l’amélioration de la prédiction des gènes eucaryotes codant pour des protéines en se focalisant spécifiquement sur les sites d’épissage 2) en exploitant des algorithmes d’intelligence artificielle (CNN et algorithmes évolutionnaires), 3) entraînés avec des données de haute qualité incluant une forte diversité d’espèces eucaryotes. Notre stratégie consiste à combiner l’ensemble des données validées avec les programmes développés afin d’améliorer la prédiction des gènes en diminuant le taux d’erreurs et éviter qu’elles ne se propagent dans les bases de données. De plus, ces travaux permettront une meilleure compréhension des organismes et de leurs mécanismes biologiques.

Gadolinium-based contrast agents (GBCAs) are indispensable tools in magnetic resonance imaging (MRI), yet their clinical use is limited by non-specific tissue accumulation, low molecular specificity, and safety concerns. Protein and peptide scaffolds provide a promising alternative because they can bind metal ions with high selectivity and enable precise molecular targeting. However, identifying short peptide motifs with optimal gadolinium (Gd ³⁺ ) coordination and high relaxivity remains a major challenge. Here, we used a machine-learning-driven peptide evolution platform, the Protein Optimization Engineering Tool (POET), to design and optimize short Gd-binding motifs that enhance longitudinal relaxivity (r ₁ ). Two algorithmic strategies were tested; motif-based and regular-expression-based representations, both were trained on an initial set of 74 twelve-amino-acid peptides derived from natural EF-hand scaffolds. Through two rounds of directed evolution and experimental screening, we identified peptides with up to a 24% increase in r ₁ ratio compared with the best natural EF-hand, and a 55% improvement in absolute r ₁ after removal of unbound Gd. Further analysis revealed that peptides exhibiting higher relaxivity generally possessed a more negative net charge and lower isoelectric point than the buffer pH, indicating stronger electrostatic stabilization of Gd ³⁺ . Sequence enrichment analysis showed that acidic and small polar residues, particularly aspartic acid, glycine, and threonine, were selectively favored during evolution, while bulky hydrophobic and basic residues were depleted. These compositional trends align with improved solubility and enhanced metal coordination. Together, these results demonstrate a generalizable framework that integrates computational evolution with biophysical screening to discover new biologically derived Gd-binding motifs. This approach provides a scalable route to engineer responsive, tunable, and biocompatible MRI contrast tags for precision imaging and molecular diagnostics.

The development of peptides for therapeutic targets or biomarkers for disease diagnosis is a challenging task in protein engineering. Current approaches are tedious, often time-consuming and require complex laboratory data due to the vast search spaces that need to be considered. In silico methods can accelerate research and substantially reduce costs. Evolutionary algorithms are a promising approach for exploring large search spaces and can facilitate the discovery of new peptides. This study presents the development and use of a new variant of the genetic-programming-based POET algorithm, called POET_Regex, where individuals are represented by a list of regular expressions. This algorithm was trained on a small curated dataset and employed to generate new peptides improving the sensitivity of peptides in magnetic resonance imaging with chemical exchange saturation transfer (CEST). The resulting model achieves a performance gain of 20% over the initial POET models and is able to predict a candidate peptide with a 58% performance increase compared to the gold-standard peptide. By combining the power of genetic programming with the flexibility of regular expressions, new peptide targets were identified that improve the sensitivity of detection by CEST. This approach provides a promising research direction for the efficient identification of peptides with therapeutic or diagnostic potential.

Background The development of peptides for therapeutic targets or biomarkers for disease diagnosis is a challenging task in protein engineering. Current approaches are tedious, often time-consuming and require complex laboratory data due to the vast search space. In silico methods can accelerate research and substantially reduce costs. Evolutionary algorithms are a promising approach for exploring large search spaces and facilitating the discovery of new peptides. Results This study presents the development and use of a variant of the initial POET algorithm, called POET_Regex, which is based on genetic programming, where individuals are represented by a list of regular expressions. The program was trained on a small curated dataset and employed to predict new peptides that can improve the problem of sensitivity in detecting peptides through magnetic resonance imaging using chemical exchange saturation transfer (CEST). The resulting model achieves a performance gain of 20% over the initial POET variant and is able to predict a candidate peptide with a 58% performance increase compared to the gold-standard peptide. Conclusions By combining the power of genetic programming with the flexibility of regular expressions, new potential peptide targets were identified to improve the sensitivity of detection by CEST. This approach provides a promising research direction for the efficient identification of peptides with therapeutic or diagnostic potential.

In fungi, the most abundant transcription factor (TF) class contains a fungal-specific ‘GAL4-like’ Zn2C6 DNA binding domain (DBD), while the second class contains another fungal-specific domain, known as ‘fungal_trans’ or middle homology domain (MHD), whose function remains largely uncharacterized. Remarkably, almost a third of MHD-containing TFs in public sequence databases apparently lack DNA binding activity, since they are not predicted to contain a DBD. Here, we reassess the domain organization of these ‘MHD-only’ proteins using an in silico error-tracking approach. In a large-scale analysis of ~17,000 MHD-only TF sequences present in all fungal phyla except Microsporidia and Cryptomycota, we show that the vast majority (>90%) result from genome annotation errors and we are able to predict a new DBD sequence for 14,261 of them. Most of these sequences correspond to a Zn2C6 domain (82%), with a small proportion of C2H2 domains (4%) found only in Dikarya. Our results contradict previous findings that the MHD-only TF are widespread in fungi. In contrast, we show that they are exceptional cases, and that the fungal-specific Zn2C6–MHD domain pair represents the canonical domain signature defining the most predominant fungal TF family. We call this family CeGAL, after the highly characterized members: Cep3, whose 3D structure is determined, and GAL4, a eukaryotic TF archetype. We believe that this will not only improve the annotation and classification of the Zn2C6 TF but will also provide critical guidance for future fungal gene regulatory network analyses.