AlphaDIA enables DIA transfer learning for feature-free proteomics
Georg Wallmann, Patricia Skowronek, Vincenth Brennsteiner, Mikhail Lebedev, Marvin Thielert, Sophia Steigerwald, Mohamed Kotb, Oscar Despard, Tim Heymann, Xie‐Xuan Zhou, Maximilian T. Strauss, Constantin Ammar, Sander Willems, Magnus Schwörer, Wen‐Feng Zeng, Matthias Mann
Abstract
The scale of data generated for mass-spectrometry-based proteomics and modern acquisition strategies poses a challenge to bioinformatic analysis. Search engines need to make optimal use of the data for biological discoveries while remaining statistically rigorous, transparent and performant. Here we present alphaDIA, a modular open-source search framework for data-independent acquisition (DIA) proteomics. We developed a feature-free identification algorithm that performs machine learning directly on the raw signal and is particularly suited for detecting patterns in data produced by time-of-flight instruments. Benchmarking demonstrates competitive identification and quantification performance. While the method supports empirical spectral libraries, we propose a search strategy named DIA transfer learning that uses fully predicted libraries. This entails continuously optimizing a deep neural network for predicting machine-specific and experiment-specific properties, enabling the generic DIA analysis of any post-translational modification. AlphaDIA provides a high performance and accessible framework running locally or in the cloud, opening DIA analysis to the community.