Rocchi, Ettore
(2026)
Computational algorithms and machine learning frameworks for omics data with focus on antimicrobial resistance, [Dissertation thesis], Alma Mater Studiorum Università di Bologna.
Dottorato di ricerca in
Scienze e tecnologie della salute, 38 Ciclo.
Documenti full-text disponibili:
![PhD_dissertation_Rocchi.pdf [thumbnail of PhD_dissertation_Rocchi.pdf]](https://amsdottorato.unibo.it/style/images/fileicons/application_pdf.png) |
Documento PDF (English)
- Accesso riservato fino a 30 Giugno 2027
- Richiede un lettore di PDF come Xpdf o Adobe Acrobat Reader
Disponibile con Licenza: Salvo eventuali più ampie autorizzazioni dell'autore, la tesi può essere liberamente consultata e può essere effettuato il salvataggio e la stampa di una copia per fini strettamente personali di studio, di ricerca e di insegnamento, con espresso divieto di qualunque utilizzo direttamente o indirettamente commerciale. Ogni altro diritto sul materiale è riservato.
Download (21MB)
| Contatta l'autore
|
Abstract
Antimicrobial resistance is a global public health threat responsible for over one million deaths annually. Timely and accurate prediction of resistance phenotypes is essential to guide empirical antibiotic therapy, yet conventional culture-based susceptibility testing requires 48-72 hours. This dissertation develops and validates computational frameworks that leverage machine learning to accelerate antimicrobial-resistancerelated clinical decision-making. In the first line of work, I address AMR prediction directly from MALDI-TOF mass spectrometry data. I introduce MaldiAMRKit, a Python library for reproducible spectral preprocessing, and combatlearn, an implementation of the ComBat algorithm family that prevents data leakage in learning pipelines. A systematic evaluation on Candida parapsilosis demonstrates that technical replicates carry isolate-specific but not population-generalizable information, establishing that training-set diversity should be prioritized over replicate depth. I further present a reproducible framework for standardized, robust evaluation of machine learning pipelines and spectral preprocessing for antimicrobial resistance predictions from MALDI-TOF mass spectra. In parallel, I develop ResPredAI, an open-source ML framework for AMR prediction in Gram-negative bloodstream infections using clinical and demographic variables. Applied to a single-center cohort, logistic regression achieves an AUROC of 0.921 for carbapenem resistance with a false omission rate of 3.5%. The framework incorporates nested cross-validation, probability calibration, and threshold optimization to meet regulatory and clinical safety requirements. Finally, I apply latent class analysis to identify infection-risk phenotypes among liver transplant recipients. Two clinically interpretable phenotypes, stratified by liver disease severity, are discovered: the high-risk phenotype exhibits 2.2-fold higher odds of 30-day post-transplant infection. All tools developed in this thesis - MaldiAMRKit, combatlearn, AMRchitecture, ResPredAI, and PhenoCluster - are released as open-source software to promote reproducibility, facilitate local clinical deployment, and enable researchers to build upon this work.
Abstract
Antimicrobial resistance is a global public health threat responsible for over one million deaths annually. Timely and accurate prediction of resistance phenotypes is essential to guide empirical antibiotic therapy, yet conventional culture-based susceptibility testing requires 48-72 hours. This dissertation develops and validates computational frameworks that leverage machine learning to accelerate antimicrobial-resistancerelated clinical decision-making. In the first line of work, I address AMR prediction directly from MALDI-TOF mass spectrometry data. I introduce MaldiAMRKit, a Python library for reproducible spectral preprocessing, and combatlearn, an implementation of the ComBat algorithm family that prevents data leakage in learning pipelines. A systematic evaluation on Candida parapsilosis demonstrates that technical replicates carry isolate-specific but not population-generalizable information, establishing that training-set diversity should be prioritized over replicate depth. I further present a reproducible framework for standardized, robust evaluation of machine learning pipelines and spectral preprocessing for antimicrobial resistance predictions from MALDI-TOF mass spectra. In parallel, I develop ResPredAI, an open-source ML framework for AMR prediction in Gram-negative bloodstream infections using clinical and demographic variables. Applied to a single-center cohort, logistic regression achieves an AUROC of 0.921 for carbapenem resistance with a false omission rate of 3.5%. The framework incorporates nested cross-validation, probability calibration, and threshold optimization to meet regulatory and clinical safety requirements. Finally, I apply latent class analysis to identify infection-risk phenotypes among liver transplant recipients. Two clinically interpretable phenotypes, stratified by liver disease severity, are discovered: the high-risk phenotype exhibits 2.2-fold higher odds of 30-day post-transplant infection. All tools developed in this thesis - MaldiAMRKit, combatlearn, AMRchitecture, ResPredAI, and PhenoCluster - are released as open-source software to promote reproducibility, facilitate local clinical deployment, and enable researchers to build upon this work.
Tipologia del documento
Tesi di dottorato
Autore
Rocchi, Ettore
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
machine learning; antimicrobial resistance; amr; MALDI-TOF; mass spectrometry
Data di discussione
15 Luglio 2026
URI
Altri metadati
Tipologia del documento
Tesi di dottorato
Autore
Rocchi, Ettore
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
machine learning; antimicrobial resistance; amr; MALDI-TOF; mass spectrometry
Data di discussione
15 Luglio 2026
URI
Gestione del documento: