Baker lab catches up to DeepMind; plus Casebia, 10X Genomics and more

A University of Washington team led by David Baker published in Science its RoseTTAFold method to predict how proteins folds based on their primary amino acid sequence. The UW team adopted the iterative feature extraction approach pioneered by the AlphaFold method from the DeepMind unit of Alphabet Inc. (NASDAQ:GOOG), which won a 2020 structure prediction competition by a wide margin, but for which details have not yet been published. The step change in protein folding prediction from both methods stands to widen the funnel of targets tackled with structure-based platforms, including newly discovered proteins from emerging pathogens. A key difference between the methods is that RoseTTAFold's network "collectively reasons" about 1-D amino acid sequences, 2-D distances between residues and 3-D spatial coordinates simultaneously, while AlphaFold processes the 1-D and 2-D information first, and incorporates the 3-D coordinates later.

Synthetic RNA-guided nucleases developed by Bayer AG (Xetra:BAYN),  CRISPR Therapeutics AG (NASDAQ:CRSP) and  Casebia Therapeutics could get around the delivery limitations imposed by the large size of Streptococcus pyogenes Cas9 without sacrificing activity or specificity. Described in Nature Communications, a synthetic nuclease based on Staphylococcus lugdunensis Cas9 had an average editing efficiency 3.1-fold higher than S. pyogenes Cas9 in a panel of 24 targets in HEK cells, with fewer unique off-targets. In non-human primates, subretinal injection of single AAV5 particles carrying the synthetic nuclease with a single guide RNA led to about 60% editing in photoreceptors, while the large size of S. pyogenes Cas9 precludes packaging into a single AAV5 genome...