We develop machine learning algorithms to distinguish disease-related mutations from neutral polymorphisms. Our foundational tools such as PhD-SNP and WS-SNPs&GO integrate sequence, evolutionary, and functional annotations (Gene Ontology) to predict variant pathogenicity. We recently updated our flagship lightweight tool, PhD-SNPg, to handle both coding and non-coding nucleotide variants using gradient boosting.
Our current efforts focus on multi-omic data integration. As part of the Italian PNRR Rare Disease project (PNRR-MR1-2022-12376067), we combine exome sequencing, transcriptomics, and DNA methylation episignatures to solve previously undiagnosed rare disease cases. We are also contributing to the ELIXIR federated infrastructure (ENIGMA and BioChef projects), implementing GA4GH standards (FAIR-compliant Variant Matching) to enable privacy-preserving cross-border analysis of genomic and imaging data. Our group contributed to a proof-of-concept study demonstrating the effectiveness of Federated Learning (FL) for the clinical classification of coding and non-coding SNVs and CNVs (Montalvo et al., Bioinformatics, 2025), supporting the adoption of secure, multi-institutional collaborations for human variant interpretation.

Understanding the relationship between structure and function is fundamental to molecular biology. Our group develops algorithms and statistical potentials for the structural alignment and assessment of RNA and proteins. In the field of RNA bioinformatics, we focus on the challenge of aligning highly flexible 3D structures to infer functional conservation. Our SARA tool implements a unit-vector approach to perform fast and accurate pairwise structural alignments of ribonucleic acids.
By merging the SARA algorithm with the T-Coffee sequence aligner, we developed SARA-Coffee. This integrated tool enables the computation of highly accurate multiple RNA structural alignments by combining tertiary structure information with sequence conservation, allowing for a more robust detection of distant evolutionary relationships.
In collaboration with the research group of Prof. Francisco Melo at the Pontificia Universidad Católica de Chile, we contributed to the development of WebRASP (http://melolab.org/webrasp). WebRASP is an all-atom knowledge-based statistical potential used to score the quality of RNA three-dimensional models and to select near-native structures from decoy sets, which is an essential step for de novo RNA structure prediction and quality assessment.

As a WP Co-Leader in the ELIXIR Human Data and Translational Research (HDTR) project, BioFolD is actively building the next generation of federated data infrastructures across Europe. By developing privacy-preserving, containerized workflows for the ENIGMA (genomics-imaging) and BioChef (hybrid workflow) platforms, we are enabling researchers to securely analyze sensitive health data across multiple countries without moving the data itself.
We extend our impact globally through the RaDAr-IT project (Patient-Oriented Data Pathways for Rare Disease Equity in Argentina). As principal investigator, BioFolD leads a cooperation between the University of Bologna and Argentinian partners (Universidad Nacional de Rosario, FundHemi) to train local professionals in FAIR data practices, variant interpretation, and responsible data governance.

We contributed to the community-driven DOME Recommendations for supervised machine learning validation in biology (Walsh et al., Nature Methods, 2021), which established reporting guidelines to improve reproducibility and assessment in biological ML studies. More recently, we co-authored the Open and Sustainable AI (OSAI) Perspective in Nature Methods (Farrell et al., 2026), providing actionable recommendations to address the reusability, reproducibility, and environmental sustainability of AI models in the life sciences. We aim to apply these best practices across our own tools and federated infrastructures, ensuring that the computational resources we develop remain open, FAIR, and environmentally sustainable.