research

my research focuses on building scientifically-grounded models for multimodal biological data, with a primary focus on oncology. in particular, developing structured multi-omics foundation models that respect the causal and statistical relationships between genomic modalities. a central theme is the incorporation of domain knowledge as structured inductive biases, rather than treating biological data as generic sequences to be modelled with off-the-shelf architectures, aligning models with the underlying biology.

in particular, i focus on three central research questions:

• how can we build models that learn rich, general-purpose representations from heterogeneous multimodal data, respecting the structure and statistical properties of each modality?
• how can machine learning methods be applied responsibly to the life sciences, enabling deeper molecular understanding of complex diseases and accelerating therapeutic discovery?
• how can scientific domain knowledge, from biophysical mechanisms to evolutionary constraints, be embedded as structured inductive biases in generative models?

theses

• reproducible knowledge distillation for graph neural networks
  l. g. stigliano, i. rekik (2023)  |  master’s thesis, imperial college london
• introduction to inverse problems and markov chain monte carlo methods
  l. g. stigliano, a. teckentrup (2021)  |  bachelor’s thesis, university of edinburgh