In recent years, data-independent acquisition (DIA) proteomics has become a widespread quantitative method, useful not only for large high-throughput screens, but also for applications that require high sensitivity, such as spatial and single cell proteomics, as well as phosphoproteomics and ubiquitinomics. Much of this progress in DIA technologies has been driven by improvements and new functionality in data processing software, such as support for ion mobility-resolved acquisition, PTM identification and multiplexing. However, wide-isolation window DIA methods still possess a fundamental limitation, that is the lack of inherent ability to confidently identify and quantify arbitrary peptidoforms. Not only can peptidoform misidentification affect protein quantities, it also holds back a range of applications of DIA proteomics, including metaproteomics and the emerging profiling of common sequence variants in population proteomics.
Here I will introduce a new neural network-based peptidoform-scoring module in our DIA-NN software suite for DIA data processing. The new module allows the use of any DIA proteomics method to confidently identify and quantify peptidoforms with statistical error-rate control. Moreover, DIA can be optimised to maximise peptidoform confidence. In fact, the high proteomic depth, characteristic of DIA methods, is largely retained when the analysis is restricted solely to peptidoform-confident identifications. The new peptidoform-scoring module in DIA-NN, therefore, addresses the long-standing limitation of DIA proteomics and paves the way for new applications of DIA proteomics that require peptidoform confidence.