How can we steer protein language models toward specific traits, without losing sequence diversity?
In our new preprint, we introduce Iterative Lookback Monte Carlo (ILMC)—a training-free sampling strategy that balances generation quality, steering constraints, and diversity.
ILMC delivers highly diverse protein designs, including variants with up to a 12°C increase in predicted melting temperature.
Full paper: https://t.co/cyOu1ZBr7I
How can we steer protein language models toward specific traits, without losing sequence diversity?
In our new preprint, we introduce Iterative Lookback Monte Carlo (ILMC)—a training-free sampling strategy that balances generation quality, steering constraints, and diversity.
We develop latent-aligned RBMs, to uncover a common representational space from cell-resolved whole-brain recordings spanning several distinct larval zebrafish individuals.
We use energy-based generative models (RBM) to design full mutational paths connecting two natural proteins of different binding specificities, and verify experimentally that intermediate sequences stay functional.