ALSEBO

Active Learning Sequence Exploration via Bayesian Optimization

ALSEBO is a Python framework for navigating protein sequence space using a closed-loop active learning strategy. A Variational Autoencoder (VAE) generates a continuous latent landscape of sequences; Bayesian Optimisation (BO) then iteratively proposes the most promising candidates for experimental testing.

VAE latent landscape
      │
      ▼
Sequence Space  (DCA · ESM · latent features)
      │
      ▼
Initial Training Set  (t-SNE / PCA + k-means diversity sampling)
      │
      ▼
Experiment  →  GPR surrogate  →  UCB acquisition  →  Next Batch
      └──────────────────────────────────────────────────┘
                     active learning loop

Contents

Indices and tables