Sarah Lewis

Sarah Lewis is a researcher in AI for science whose work spans protein design, protein dynamics, and materials discovery. She holds mathematics degrees from the University of Cambridge and a neuroscience PhD from Imperial College London; in 2020, she joined Microsoft's AI Residency and worked on holographic storage, missing-data models, and compiler reinforcement learning. She coauthored FrameFlow for protein-backbone generation and motif scaffolding, BioEmu for protein ensembles, and a 2025 paper on MatterGen for crystalline materials. As of October 2026, she is a Senior Research Engineer at Microsoft Research Cambridge, applying machine learning to scientific problems such as drug design.

Now at: Microsoft Research
Works: 5
Citations: 858
h-index: 3

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2020

Microsoft AI Residency

She holds bachelor's and master's degrees in mathematics from the University of Cambridge and a PhD in neuroscience from Imperial College London. Before Microsoft, she works as an algorithm developer and technical lead in multidisciplinary contract R&D. In 2020, she joins Microsoft's AI Residency and works on holographic-storage data density, probabilistic models for missing data, and reinforcement learning for an optimizing compiler.
2023

FrameFlow protein generation

In 2023, she coauthors FrameFlow, which uses SE(3) flow matching to generate protein backbones. The paper reports five times fewer sampling steps and twofold better designability than FrameDiff.
January 2024

FrameFlow motif scaffolding

In January 2024, she coauthors an extension of FrameFlow for motif scaffolding, with motif amortization and motif guidance. On a benchmark of 24 biologically meaningful motifs, the method achieves 2.5 times more designable and unique scaffolds than state-of-the-art methods.
2024

BioEmu protein ensembles

In 2024, she coauthors BioEmu, a deep learning model that samples thousands of independent structures from a protein sequence each hour. The model predicts equilibrium ensembles and captures functional changes such as cryptic-pocket formation, local unfolding, and domain rearrangements.
January 2025

MatterGen materials design

In January 2025, she coauthors a Nature paper introducing MatterGen, a diffusion model for designing crystalline materials across the periodic table.
February 2025

Open-sourcing BioEmu-1

In February 2025, she and the Microsoft Research AI for Science team make BioEmu-1 open source. The deep learning model generates thousands of protein structures per hour on one GPU, helping researchers study the different structures proteins can take.
As of October 2026

Research at Microsoft Cambridge

As of October 2026, she is a Senior Research Engineer at Microsoft Research Cambridge, applying machine learning to scientific problems such as drug design.