About
Hi, I’m Giovanni. I’m a biologist who became a computational scientist, and I work where biomedical knowledge representation meets deep learning on graphs.
My doctoral research asks a deceptively simple question: how much biological semantics is worth formalising in a data representation, and what does that formalisation actually buy a learning system? To answer it I build ontology-grounded knowledge graphs that carry per-sample multi-omic measurements, and heterogeneous graph neural networks that learn over them, for cancer subtype classification and drug–target discovery.
I came to computation from molecular genetics and nutrition science, and I still think that background is the useful part: the hard problems in this field are rarely about the architecture, and almost always about what the data is allowed to mean.
- Position PhD candidate in Computational and Quantitative Biology (cycle XXXIX)
- Affiliation DIETI, University of Naples Federico II
- Supervisor Prof. Antonio Maria Rinaldi
- Previously Visiting researcher, Computer Laboratory, University of Cambridge (2025)
- ORCID 0009-0002-8395-0724
- Elsewhere Scholar · GitHub · LinkTree
Doctoral thesis
From Trans-Omic Networks to Knowledge Graphs: Encoding and Exploiting Biological Semantics for Multi-Omics Data Integration with Graph Neural Networks
In multi-omics the bottleneck is no longer measurement, it is representation. Every integrative study rebuilds its own encoding, and existing biomedical knowledge graphs keep per-sample quantitative vectors in external matrices, breaking the link between an entity and its measurement at query time.
The thesis builds one reusable substrate where an ontological backbone and dense per-sample multi-omic profiles are stored together and retrieved through a single query, then puts that substrate to work three ways: joint querying of knowledge and measurement, discovery of new biological relations, and molecular subtype classification. The organising axis is the degree of semantic formality of the prior, from implicit trans-omic networks, to shallow molecular priors, to a formal multi-scale knowledge graph.
The finding is a careful one, and it is the point of the work: the quantitative signal dominates, and formal semantics contributes selectively, trading marginal accuracy for robustness under feature scarcity, stability across data splits and mechanistic interpretability, which are the properties that matter when complete omic panels are unavailable, as they usually are in the clinic.
Research interests
- Multi-omics integration with ontology-grounded knowledge graphs
- Heterogeneous and multi-relational graph neural networks for classification and link prediction
- Knowledge graph embedding for drug repurposing, including across species
- Interpretable and calibrated models for clinical decision support
- LLMs and retrieval over biomedical literature and structured knowledge
- Nutrigenetics and genotype-informed personalised nutrition
A fuller account is on the research page, and the record of published work on the publications page.
“A path is made by walking on it.” ― Chuang Tzu