FILE № AA–2026.01PERSONAL INDEXREV. 2026-05-19
ANDREAANELLI.
001NAMEAndrea Anelli, PhD
002ROLEML Research Scientist
003FIELDAtomistic simulation · Drug discovery
004BASEDBasel, Switzerland
005AFFILIATIONRoche · pCMC · D2AI
006INTERESTSPhotography · Music Production · Gemstones · Automation
ABSTRACT —
I design machine learning architectures for atomistic simulation and deploy them on the problems that define modern drug discovery — crystal structure prediction, preformulation automation, and structure–property learning across scales.
0
Publications
0+
Citations (Scholar)
0
h-index
0
i10-index
§ 01
All research →Selected work
01
APPA · Agentic Preformulation Pathway Assistant
An LLM-based agentic system that couples scientific reasoning with experimental databases and ML predictions — replacing weeks of manual preformulation evidence gathering with a single autonomous loop.
0citations
02Divide-and-Conquer Potentials for Scalable Atomistic Simulation
A committee-of-experts architecture where specialised ML potentials are weighted by their expertise — yielding more accurate predictions than monolithic models at essentially no additional cost.
0citations
03The Seventh Blind Test of Crystal Structure Prediction
Community benchmark of 21 groups generating crystal structures for pharmaceutical targets. Showed that ML-guided generation now reliably finds experimental polymorphs for rigid molecules — a watershed for industrial CSP.
0citations
§ 02
A research arc
From model architectures, to the physical properties they learn, to the industrial tools they power, to the agentic systems that orchestrate them — my work reads bottom-up, connecting atoms to autonomous reasoning.
01
ML Potential Architectures
Building the models that simulate atoms.
02
Structure–Property Learning
From atomic coordinates to macroscopic behaviour.
03
Crystal Structure Prediction
Which polymorph will a drug molecule form?
04
Agentic AI for Drug Discovery
Autonomous systems that reason like scientists.