Notes from the design loop.
Protein engineering methods, sequence co-variation intuitions, and what we have learned from watching wet-lab teams use generative search in real programs.
How to Read a Fitness Landscape Before You Design a Single Variant
Before generating candidates, understanding the shape of the fitness landscape tells you whether single-residue walks will work or whether you need combinatorial diversity.
Nanobody vs. Antibody CDR Engineering: Where Generative Search Helps Most
The smaller scaffold of nanobodies makes sequence space coverage different in ways that affect how you should set up a generative design run.
Thermostability and Activity: Why You Rarely Get Both, and When You Can
The classic trade-off between thermal stability and catalytic activity has structural roots. Generative models can sometimes find sequences that thread the needle, but not always.
Retrospective Validation: What Rank Correlation on Your Own Data Actually Tells You
Before committing to any computational design tool, running a retrospective on held-out data is the only honest quality check. Here is how to interpret the result.
A Primer on Sequence Co-variation for Protein Engineers Who Are Not ML Researchers
Evolutionary co-variation in protein sequences encodes structural and functional constraints. Understanding it helps you interpret why a model ranks certain mutations favorably.
Broadening Substrate Specificity in Industrial Enzymes: A Case Study Framing
When a specialty chemical process needs an enzyme to accept a slightly different substrate, the design problem is narrower than therapeutic engineering but the stakes are equally high.
Closing the Wet-Lab Dry-Lab Loop: How Fast Should the Feedback Cycle Be?
The latency between an assay result and the next computational design run determines how many iterations you can afford. We think about this as a scheduling problem.
What UniProt Coverage Means for Your Protein Family (And When It Runs Out)
A generative model trained on public sequence databases performs well when your target protein family has broad evolutionary representation. Here is how to check before you start.
Why We Started Proteinvue
Elena and Marcus describe the specific moment in a protein engineering lab that made it obvious a computational shortlist tool had to exist.
MSA Depth and Model Confidence: How Many Homologs Do You Actually Need?
Multiple sequence alignment depth is a proxy for how much evolutionary signal a model can draw on. Shallow MSAs degrade confidence in novel regions of sequence space.
What Do We Mean by Fitness Objective, and How Do You Define Yours?
The single most important input to a generative protein design run is a clear definition of what you are optimizing for. Getting this wrong is more expensive than any model limitation.