Dr. Hayden Metsky: ADAPT for Large-scale Viral Detection

Dr. Hayden Metsky begins by introducing the ADAPT method for doing large-scale detection of viruses. ADAPT is a computational method that aids the design of CRISPR-based viral testing. He then discusses the motivation for ADAPT and how it relates to his previous works like CATCH. In comparing ADAPT to other work, Dr. Metsky discusses, for instance, differences between CRISPR-based testing and more traditional testing like qPCR. In discussing the challenges of designing diagnostic tests and detection assays, Dr. Metsky then describes how he breaks these challenges into three different components. 

Dr. Metsky goes on to talk about how they designed an assay based around the Cas13 enzyme. He describes how they used this approach for targeting viruses and explains how they designed a large-scale library of 20,000 pairs of target viral sequences and guide RNAs. He then explains how they used this library as training data for a machine learning model. He also explains his thought process of designing and training the convolutional neural network model that they ended up using for predicting how well the guide RNAs would work. 

As our conversation continues, Dr. Metsky points out an interesting observation that his team made while working on this project. He points out that it could be the case that it may not be best to only design diagnostics around a highly or universally conserved region. He explains that taking into account other considerations, like how well the diagnostic technology works for a particular target sequence, may produce even better results. He also points out how it can be really challenging to only consider the highly conserved or totally conserved regions because those regions are going to be the most likely to be shared by other viruses or organisms which induce false positives in tests. Dr. Metsky explains his thought process for how you take a problem like this, figure out the characteristics of the problem, and match it well to a closely related problem or other scientific works, explaining the process of figuring out how to optimize the final objective function in ADAPT. Final topics include a discussion on the speed of ADAPT and the availability of the software. 

To learn more about ADAPT, you can read the ADAPT manuscript at https://www.biorxiv.org/content/10.1101/2020.11.28.401877v2 or visit the software page at https://github.com/broadinstitute/adapt