Previously mentioned, we experimented with the idea before the current state of machine learning, with a kind of simulation more akin to 'digital twins'. The ML method would have been useful to add.
Deep Learning Takes on Synthetic Biology
The Harvard Gazette By Lindsay Brownell October 7, 2020
Two teams of scientists from Harvard University and the Massachusetts Institute of Technology have developed machine learning algorithms that can analyze RNA-based "toehold switch" molecular sequences and predict which will reliably sense and respond to a desired target sequence. The researchers first designed and synthesized a massive toehold switch dataset, which Harvard's Alex Garruss said "enables the use of advanced machine learning techniques for identifying and understanding useful switches for immediate downstream applications and future design." One team trained an algorithm to analyze switches as two-dimensional images of base-pair possibilities, and then to identify patterns signaling whether a given image would be a good or a bad toehold via an interpretation process called Visualizing Secondary Structure Saliency Maps. The second team tackled the challenge with orthogonal techniques using two distinct deep learning architectures. Their Sequence-based Toehold Optimization and Redesign Model and Nucleic Acid Speech platforms enable the rapid design and optimizing of synthetic biology components. ... "
No comments:
Post a Comment