Protein ML Talks
Recent talks applying the latest protein-ML models for design under the framework of p(sequence, structure, function).
Generative Models for Protein Design
- Training data of generative models from evolutionary and PSI-BLAST search
- ESM pseudo-likelihoods are inversely correlated with enzyme thermostability
- Autoregressive, bidirectional, and graph-based masked-token prediction training tasks
- Conditioning by adaptive sampling