Corel

A Continuous Relaxation for Discrete Bayesian Optimization

DISCLAIMER: This repository is currently under development and subject to significant changes until an official release.

Setup

Create dedicated virtual environment:

conda env create -f environment.yml

Activate corel-env

Setting up Poli

git clone poli && cd ./poli/
pip install -e .

Run Experiments

To replicate the RFP results run

python experiments/run_cold_warm_start_experiments_rfp_bo.py

Datasets

RFP

The original RFP dataset was used as listed in LaMBO reference work. Known sequence structures were extracted and aligned with clustalo binaries (see Clustal Omega v. 1.2.4 64-bit Linux binary) -> MSA. The aligned MSA was used with HMMER to build a sequence alignment for VAE training using the MPI Bioinformatics Toolkit (see Zimmermann et al.).:

• Reference hmmsearch against alphafold_uniprot50DB (default).
• E-value cutoff set to 1 to obtain also distant alignments.
• columns were filtered gap 20%

See Dataset readme for details and citations.

GFP

Contained within the poli package, references provided there.