Extended Data Fig. 10: Entrapment validation of end-to-end transfer learning across four iterations.
From: AlphaDIA enables DIA transfer learning for feature-free proteomics
a, Overview of the validation workflow. A Human and Arabidopsis fasta file digest was used for fully predicted library search. All identified precursors at 1% FDR were subsequently used for DIA transfer learning, including false positive Arabidopsis identifications. This process was repeated twice, using the transfer learned deep-learning model for library prediction. b, Total unique identified precursors across six replicates. Precursors mapping to both species, including leucine and isoleucine pairs were removed. c, Total unique identified protein groups. d, Entrapment FDR given as the percentage of false positive Arabidopsis identifications. e, MS2 spectral angle for precursors before and after transfer learning. Median spectral angle is shown for each plot (nHuman=283,383, nArabidopsis = 234, boxplot according to Methods). f, Retention time deviation in seconds before and after transfer learning. The median retention time deviation is shown across three replicates (nHuman=283,383, nArabidopsis = 234, boxplot according to Methods). g, Predicted vs observed retention time following transfer learning. False positive Arabidopsis identifications are highlighted.
