ProteoTorch¶
A Python package for deep learning and fast machine learning analysis of MS/MS database search results.
Key Features¶
ProteoTorch accepts as input a Percolator INput (PIN) file containing target/decoy PSM features. By default, several iterations of deep semi-supervised learning are then performed to classify target and decoy PSMs, and the output PSM scores are recalibrated using the resulting learned parameters.
ProteoTorch provides the following semi-supervised machine learning classifiers:
Deep neural networks
Fast Linear SVMs using the L2-SVM-MFN algorithm (equivalent to the recently sped-up Percolator algorithm)
Linear SVMs using the TRON algorithm (equivalent to these Percolator speedups)
Linear Discriminant Analysis (+ Gaussian Mixture Models, in development)
Support to easily swap in any supervised classifier implemented in Python which follow the design of scikit-learn clf object instances, with training function fit and testing function decision_function
Additional Features¶
ProteoTorch provides an ultrafast q-value library (heavily optimized for Python), plotting tools to benchmark/compare MS/MS post-processor results, and an easy-to-use Python API for the MS/MS semi-supervised learning algorithm (with cross-validation) originally implemented in the C++ package Percolator.