PhD Researcher · University of Bologna
Physics background, biomedical mission.
Building interpretable machine learning pipelines for clinical microbiology and multi-omics data.
I develop computational methods to predict antimicrobial resistance, discover patient phenotypes, and make sense of high-dimensional omics data. My work spans MALDI-TOF mass spectrometry, multi-omics integration, genomics, and metagenomics, with a focus on interpretability and clinical impact.
I trained as a physicist and moved into biomedical data science to apply quantitative methods to problems with direct clinical impact. My current work focuses on antimicrobial resistance prediction from MALDI-TOF mass spectrometry, multi-centre data harmonisation, computational patient phenotyping, and computational genomics.
I work within the Physics4MedicineLab, part of the Multi-Omics and Health-Care Data Analytics Unit at Sant'Orsola Hospital in Bologna. My research combines machine learning with clinical microbiology, pharmacokinetics, and genomics. I care about making models that are not only accurate but interpretable, because a prediction without an explanation is just a well-dressed guess.
I write Python, build pipelines, publish open-source tools, and contribute to interdisciplinary projects. The through-line is a commitment to reproducibility and a mild obsession with doing things properly.
In practice, my days involve writing Python scripts until something either converges or breaks, reading papers about problems I didn't know existed until last week, preprocessing mass spectra that have strong opinions about baseline correction, and discussing model outputs with clinical collaborators, figuring out together when the data is telling a story we hadn't expected.
There's also a fair amount of staring at loss curves, debugging pipelines that worked yesterday, and writing documentation future-me will thank present-me for. The ratio of thinking to typing is higher than most people expect. The ratio of coffee to output is best left unquantified.
Institutions and groups I currently work with on research projects.
Four threads I am currently pulling, at the intersection of applied physics, machine learning, and clinical data.
Machine learning on mass spectra and clinical data to anticipate antimicrobial resistance ahead of culture-based diagnostics, with cross-site harmonisation and generative modelling extending the pipeline beyond single-instrument settings.
MALDI-TOF · supervised & generative learning · cross-site harmonisation
Stratification of infectious risk in fragile populations such as transplant recipients, and surveillance of circulating pathogens through metagenomic monitoring and computational phenotyping.
patient phenotyping · survival & multi-state models · metagenomic surveillance
Discovery and interpretation of structural and somatic variants from short- and long-read sequencing, in clinically relevant genomic contexts.
structural variants · somatic calling · long-read sequencing
Open-source tools and reusable methods built around specific biomedical questions, designed to be reproducible, well-documented, and useful beyond their original project.
bioinformatic tools · open-source software · reproducible pipelines
Open-source tools and research pipelines. A selection below; the full catalog lives on the projects page.
A Python ecosystem for end-to-end clinical AMR pipelines on MALDI-TOF spectra.
Three sklearn-compatible packages that chain into a single workflow: preprocess with MaldiAMRKit, harmonise across batches and sites with MaldiBatchKit, classify with MaldiDeepKit. Designed to be modular, reproducible, and clinically deployable.
AI model to predict resistances in Gram-negative bloodstream infections.
CLI-based ML framework with nested cross-validation, nine model architectures, probability calibration, and threshold optimization for clinical decision-making. Configured via a single .ini file. Published in npj Digital Medicine.
Unsupervised clinical phenotype discovery with survival and multi-state modeling.
Tools for identifying clinically meaningful patient subgroups from heterogeneous cohorts, combining unsupervised clustering with survival analysis and multi-state trajectory modelling for prognostic interpretation.
A few recent highlights. A curated list with BibTeX lives on the publications page; the complete record is on Google Scholar and Scopus.
Combining mass spectrometry and machine learning models for predicting Klebsiella pneumoniae antimicrobial resistance: a multicenter experience from clinical isolates in Italy
Artificial Intelligence model to predict resistances in Gram-negative bloodstream infections
CATS: a bioinformatic tool for automated Cas9 nucleases activity comparison in clinically relevant contexts
A physics training that gradually re-oriented itself toward biomedical questions.
University of Bologna
Supervisor: Prof. Gastone Castellani
Visiting researcher · 6 months
Max Planck Institute of Biochemistry, Munich, Germany
Machine Learning & Systems Biology group - Prof. Karsten Borgwardt
University of Bologna
University of Bologna
The day-to-day toolbox - languages, libraries, and workflow managers I rely on.
I am always happy to talk about research collaborations and open-source development, but also about ideas, methods, and topics I have not explored yet. I am as interested in learning something new from someone as in starting something new together. If you have an idea, a question, or a dataset that misbehaves, I would be glad to hear about it.
Based in Bologna, Italy