Benchmarking Methods for Protein Design using Variable Length Flow Matching and Diffusion Models
Dnr:
NAISS 2026/4-1630
Type:
NAISS Small
Principal Investigator:
Hedwig Nordlinder
Affiliation:
Karolinska Institutet
Start Date:
2026-09-15
End Date:
2027-04-01
Primary Classification:
10203: Bioinformatics (Computational Biology) (Applications at 10610)
Webpage:
Allocation
- Arrhenius GPU at NAISS: 500 GPU-h/month
- Arrhenius Flash at NAISS: 500 GiB
Abstract
Current protein design models require the length of the protein to be specified in advance. Recently, we developed a new flow matching framework, Branching Flows (https://arxiv.org/abs/2511.09465), that allows the number of elements in the flow state to change during design. This project will further explore variable length protein design using this and other approaches, training new models and performing benchmarks.
The main supervisor is Dr. Benjamin Murrell, Principal Researcher at Karolinska Institutet