A curated catalog of the software I build and maintain. Most are Python packages,
all are open-source, and each was written to solve a problem I actually ran into.
Pull requests, issues, and feedback are welcome.
A Python ecosystem for MALDI-TOF spectral processing and analysis in antimicrobial resistance research. Three sklearn-compatible packages that chain into an end-to-end clinical AMR pipeline.
pip install MaldiSuite
MaldiAMRKit
Preprocess
Smoothing, baseline correction, peak detection (including persistent homology), alignment strategies, and flexible binning. Parallelised, sklearn-integrated, opinion-free about which method you should use.
Batch-effect correction tailored to MALDI-TOF data, with leakage-safe cross-validation handling and diagnostics that quantify residual structure after correction. Multi-site by design.
Deep learning models for MALDI-TOF spectra, with calibrated outputs and clinically oriented evaluation (decision thresholds, cost-sensitive metrics, cross-site validation).
Self-contained ML frameworks built around a specific clinical question. Each one
ships as an installable package with a single configuration file, runs end-to-end
from raw inputs to reproducible results and figures, and has been used in peer-reviewed
work.
A reproducible ML framework for predicting antimicrobial resistance from clinical and microbiology data.
Built around nested cross-validation with hyperparameter selection, nine model architectures (linear, tree-based, kernel methods, neural networks), probability calibration, and decision-threshold optimization. The whole pipeline is configured via a single .ini file and produces reproducible results, including calibration plots, performance tables, and SHAP-based interpretability reports.
Used in: Bonazzetti C, Rocchi E, et al.npj Digital Medicine, 2025
@article{Bonazzetti_2025,
title={Artificial Intelligence model to predict resistances in Gram-negative bloodstream infections},
volume={8},
ISSN={2398-6352},
url={http://dx.doi.org/10.1038/s41746-025-01696-x},
DOI={10.1038/s41746-025-01696-x},
number={1},
journal={npj Digital Medicine},
publisher={Springer Science and Business Media LLC},
author={Bonazzetti, Cecilia and Rocchi, Ettore and Toschi, Alice and Derus, Nicolas Riccardo and Sala, Claudia and Pascale, Renato and Rinaldi, Matteo and Campoli, Caterina and Pasquini, Zeno Adrien Igor and Tazza, Beatrice and Amicucci, Armando and Gatti, Milo and Ambretti, Simone and Viale, Pierluigi and Castellani, Gastone and Giannella, Maddalena},
year={2025},
month=May
}
An unsupervised framework for clinical phenotype discovery with survival and multi-state outcomes.
Combines unsupervised clustering with survival analysis and multi-state trajectory modelling to surface clinically meaningful patient subgroups from heterogeneous cohorts. Designed for prognostic interpretation: every cluster comes with hazard ratios, transition probabilities, statistical reports, and graphical summaries.
Built for: heterogeneous patient cohorts where outcomes are time-to-event or multi-state
pip install phenocluster
Pythonclusteringsurvivalmulti-statephenotyping
Standalone tools
Single-purpose libraries & bioinformatic tools
Focused tools that solve one problem well. Some plug into the frameworks above,
others stand on their own and are used by collaborators across labs.
Integrates Johnson, Fortin (neuroComBat), and Chen (CovBat) harmonization methods into scikit-learn pipelines, with leakage-safe cross-validation handling. Plugs into existing workflows without breaking the scikit-learn API contract.
Nested cross-validation with calibration, threshold optimization, and statistical tests.
A toolkit for rigorous model evaluation: nested CV with hyperparameter selection kept inside the inner loop, probability calibration, decision-threshold tuning, and built-in statistical comparisons between models.
Automated Cas9 PAM-compatibility comparison with ClinVar integration.
A bioinformatic tool for comparing Cas9 nucleases across clinically relevant genomic contexts. Detects overlapping PAM sites between variants and identifies allele-specific targets arising from pathogenic mutations. Published in Frontiers in Genome Editing (2025).
@article{Rocchi_2025,
title={CATS: a bioinformatic tool for automated Cas9 nucleases activity comparison in clinically relevant contexts},
volume={7},
ISSN={2673-3439},
url={http://dx.doi.org/10.3389/fgeed.2025.1571023},
DOI={10.3389/fgeed.2025.1571023},
journal={Frontiers in Genome Editing},
publisher={Frontiers Media SA},
author={Rocchi, Ettore and Magnani, Federico and Castellani, Gastone and Carusillo, Antonio and Tarozzi, Martina},
year={2025},
month=July
}
Broken-stick model extension for metagenomic simulation.
Extends the CAMISIM metagenomic simulator with a broken-stick abundance model and a configurable number of strains, producing synthetic communities with controlled relative-abundance distributions for benchmarking metagenomic pipelines.
Interactive toolkit for sarcomeric morphometry and myofibril linearity analysis.
Combines image analysis with circular statistics and Monte Carlo simulations to quantify sarcomere geometry (aspect ratio, curviness, shape index) and myofibril alignment in skeletal-muscle micrographs.