Original title: Advances in Machine Learning for Directed Evolution: A Five-Year Retrospective
Bruce J. Wittmann · September 2026
Scientists use directed evolution to breed useful proteins in a lab. Over the last five-plus years, machine learning transformed many areas of protein engineering, but not directed evolution. The author argues this happens because computer scientists and lab scientists want different things. Machine learning researchers try to find one perfect, optimal protein. Real-world experimentalists just need a protein that is good enough while dealing with limits on their time and money. Because of this gap, almost all current computer-guided methods ignore the real-world cost of DNA synthesis, which means building artificial genetic material. This flaw makes many methods impractical no matter how smart the models are. Fortunately, a few recent projects avoid this mistake, showing these two goals can successfully work together.
Imagine planning a dream vacation where you choose the absolute best flights, hotels, and activities without checking your budget or how much time you actually have. When it's time to book, you realize the plan is too expensive or long to pull off. In this paper, machine learning for protein engineering is like that overambitious vacation planning—it aims for the perfect protein but ignores real costs like synthesizing DNA. As a result, the suggested proteins often can't be made or tested in actual labs.
Protein engineers might waste resources if they rely on AI methods that ignore real-world costs. This disconnect means theoretical advances won't help in the lab unless they match practical needs. Aligning AI goals with constraints could make protein design actually useful.
The abstract does not present new experimental data or propose a specific method to solve the disconnect. It is a discussion piece that critiques current approaches and highlights exceptions based on the author's perspective. The claims about neglecting costs are not backed by a systematic review in this abstract.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Advances in Machine Learning for Directed Evolution: A Five-Year Retrospective — Bruce J. Wittmann
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.