Journal: Nature Communications
Article Title: A machine learning toolkit for genetic engineering attribution to facilitate biosecurity
doi: 10.1038/s41467-020-19612-0
Figure Lengend Snippet: a Two-dimensional tSNE of the pre-logit hidden layer of deteRNNt for the validation set. Colored by probability assigned to the true lab. An interactive 3d visualization is available at http://papers.altlabs.tech/visualize-hiddens.html . b Scanning mask of 10 N’s across linear sequence ( x axis) of Chris Voigt lab plasmid pCI-YFP, (Genbank JQ394803.1 ) on the brink of predicting Baojun Wang Lab vs. Voigt Lab. At each x position is shown the probabilities given by softmax on the mean of the logits from all the sequences which include that position masked with N. Plasmid schematic is shown below. c The scanning 10-N mask analysis from ( b ) applied to Edward Boyden Lab Plasmid pAAV-Syn-SomArchon (Addgene # 126941 ). The second-most likely lab predicted by deteRNNt on this plasmid is shown in red, with plasmid schematic above. d Predicting lab-of-origin from subsequences of varying lengths K scanned across the linear sequence. At each x , the probabilities are given by softmax of the mean of logits from subsequences which include that position in the window. Color indicates subsequence K -mer length. Linear sequence position is given by the x axis and shared with ( b ) with the plasmid schematic above. e Custom-designed gene-drive plasmid derived from an Omar Akbari Aedes aegypti germline-Cas9 gene-drive backbone (AAEL010097-Cas9, Addgene # 100707 ) carrying a payload of Cas9-dsRed, a guide cassette, and SomArchon from the Boyden plasmid pAAV-Syn-SomArchon from ( c ), with scanning subsequence window as in ( d ) and K = 1024. Predicted probabilities for Omar Akbari (red) and Boyden (blue) are shown, along with the correspondingly colored plasmid schematic. All unlabeled regions are from the Akbari vector. The right and left homology arms shown in gray were introduced by our design and come from neither lab’s material, and the U6a and U6c were from Akbari lab material (Addgene # 117221 and # 117223 , respectively) but introduced into the backbone by our design (gray outlined in red).
Article Snippet: Colored by probability assigned to the true lab. An interactive 3d visualization is available at http://papers.altlabs.tech/visualize-hiddens.html . b Scanning mask of 10 N’s across linear sequence ( x axis) of Chris Voigt lab plasmid pCI-YFP, (Genbank JQ394803.1 ) on the brink of predicting Baojun Wang Lab vs. Voigt Lab. At each x position is shown the probabilities given by softmax on the mean of the logits from all the sequences which include that position masked with N. Plasmid schematic is shown below. c The scanning 10-N mask analysis from ( b ) applied to Edward Boyden Lab Plasmid pAAV-Syn-SomArchon (Addgene # 126941 ).
Techniques: Biomarker Discovery, Sequencing, Plasmid Preparation, Derivative Assay
Addgene inc is a verified supplier 
