Arino, Silvia (2025) Harnessing Deep Learning for Antimicrobial Peptide Design: A Journey from Computational Models to Functionalized Membranes. [Tesi di dottorato]

Documento PDF Arino_Silvia_XXXVII_PhD.pdf Visibile a [TBR] Amministratori dell'archivio Download (4MB) | Richiedi una copia

Abstract

Advances in tools and methods from Artificial Intelligence (AI) applied to peptide and protein design has strongly propelled the design of novel biomolecules with new shapes and desired molecular functions, with applications ranging from diagnostics, medicine, catalysis and material science. Within the area of AI, Deep Learning (DL) is a core technology thanks to its ability to independently learn and extract meaningful patterns from complex data. For these features, DL algorithms represent useful tools in the context of computing and are now widely applied in a variety of research fields. In particular, they offer several advantages over conventional methods in the field of peptide design. Driven by the fascinating application of AI tools in guiding future discoveries in life science, this Ph.D. thesis has been focused on the use of DL approaches for the development of novel peptide sequences with antimicrobial activity to be used in real world applications. This goal was pursued by addressing two main aspects: the application of DL algorithms to design antimicrobial peptide sequences and subsequent validation by wet-lab experiments (described in Section 1) and immobilization of the most active peptide sequence on Nylon membranes for the development of new materials with antimicrobial behaviors (described in Section 2). Section 1 :The application of two DL algorithms, HydrAMP and Amplify, allowed the design and selection of three peptide sequences, namely AMP1, AMP and AMP3 with in silico predicted antimicrobial activity. A comprehensive validation of the computational design results was performed by wet lab experiments, with a first focus on the determination of the peptide antimicrobial activity on different bacterial strains. In details, the Minimum Inhibitory Concentrations (MIC) against various bacterial strains (e.g., E. coli, S. aureus), was determined by using the two-fold broth microdilution method. AMP3 emerged as the most active among the three sequences, both against E. coli and P. aeruginosa, which is notorious for its antimicrobial resistance, exhibiting MIC values of 15 μM and 125 μM, respectively. A detailed biophysical characterization on model membranes, by combining differential scanning calorimetry, circular dichroism, and fluorescence spectroscopy, was carried out on AMP3. This revealed that the efficacy of AMP3 peptide relies on its ability to efficiently integrate into the lipid bilayers, as highlighted by the very high mole fraction partition constant value (Kx). Section 2: The peptide sequences developed were assayed for real-world applications by investigating their behaviors upon immobilization on Nylon membranes. To this end, peptide sequences were incorporated into the polymer matrix by a grafting through strategy. A detailed physico-chemical analysis, by DSC, WAXS and UV-Vis spectroscopy, confirmed the effective incorporation of the peptide without compromising the semi-crystalline structure of Nylon 6,6. Next, Peptide-Nylon membranes were produced by electrospinning and phase inversion approaches. The membranes produced by electrospinning showed a nanofiber like structure, high porosity and improved hydrophilic properties, making them particularly suitable for antimicrobial applications. Interestingly, these peptide-membrane composite assayed for their activity against E. coli, caused a significant reduction of bacterial viability. In conclusion, the research activities carried out and the results obtained in this PhD project demonstrate the useful application of DL algorithms to design novel and active AMPs. Further, the combined approach of computational predictions and experimental implementation allowed the construction of multifunctional materials with potential applications in different fields.

Downloads

Actions (login required)

Modifica documento