Daniele Pessina
I build computational models at the interface of machine learning and biology: differentiable, uncertainty-aware tools that turn messy experimental data into decisions.
Currently
Virtual human avatars of toxicity
I develop in-silico models that predict how drugs are absorbed, distributed, metabolized, and become toxic across human organs, part of a consortium effort to make drug-safety prediction accurate enough to reduce, and ultimately replace, animal testing.
Doctoral research
Alongside Deep Origin, I'm finishing my PhD at the Sargent Centre for Process Systems Engineering, Imperial College London, supervised by Dr. Maria Papathanasiou and Prof. Jerry Heng, funded through a CASE studentship with AstraZeneca.
My thesis develops differentiable, uncertainty-aware models for template-induced biomolecule crystallisation and pharmacokinetics which combine mechanistic simulation, machine learning, and probabilistic methods to guide experimental design and accelerate process development.
What I work on
- Time-series modelling in data-scarce regimes — neural ODEs, transformers, and autoencoders with transfer and reinforcement learning.
- Probabilistic programming & uncertainty quantification for mechanistic, hybrid, and black-box models, enabling robust estimation under uncertainty.
- Differentiable programming for GPU-accelerated, autodiff-compatible simulation in Julia and JAX.
- Hybrid modelling & ML-assisted sensitivity analysis of large, non-linear pharmacokinetic systems.