Fig. 1: Overview of the AlphaPeptDeep framework.
From: AlphaPeptDeep: a modular deep learning framework to predict peptide properties for proteomics
a Measured peptide properties are encoded with the respective amino acid sequences and used to train a network in AlphaPeptDeep (left). Once a model is trained, it can be used on arbitrary sets of peptide sequences to predict the property of interest. This can then improve the sensitivity and accuracy of peptide identification. b The AlphaPeptDeep framework reads and embeds the peptide sequences of interest. Its components include the build functionality in which the model can build. Meta embedding refers to the embedding of meta information such as precursor charge states, collisional energies, instrument types, and other non-sequential inputs. It is then trained, saved and used to predict the property of interest. The dial represents the different standard properties that can be predicted (RT retention time, CCS collision cross section, MS2 intensities of fragment spectra). Custom refers to any other peptide property of interest. The lower part lists aspects of the functionalities in more detail. NN neural network, LSTM long short-term memory, CNN convolutional neural network, GRU gated recurrent unit, API application programming interface, PTM post-translational modification.
