DeepSec

DeepSec predicts secreted proteins from protein sequences to identify secretome components in human body fluids for proteomics analyses and biomarker discovery.

Key Features:

• Sequence-based end-to-end approach: An end-to-end sequence-based model operates directly on protein amino-acid sequences to predict secretion.
• Convolutional Neural Network (CNN): A CNN component learns abstract local features from protein sequences.
• Bidirectional Gated Recurrent Unit (Bi-GRU) and fully connected layer: A Bi-GRU followed by a fully connected layer models sequential dependencies and performs classification.
• Integration of architectures: The combination of CNN and Bi-GRU captures complex sequence patterns and dependencies.
• Evaluation across fluids: Validated on datasets from 12 different human body fluids.
• Performance metrics: Reported average area under the ROC curve (AUC) between 0.85 and 0.94 across testing datasets for each fluid type.
• Addresses proteomics challenges: Targets variability arising from large protein numbers, diverse post-translational modifications, and technical limitations of mass spectrometry.
• Biomarker discovery case study: Applied to kidney cancer genomics data from The Cancer Genome Atlas (TCGA) to identify 104 candidate marker proteins.

Scientific Applications:

• Secretome identification: Identification of secreted proteins across major human body fluids.
• Biomarker discovery: Discovery of candidate protein biomarkers, exemplified by a TCGA kidney cancer case study yielding 104 possible marker proteins.
• Comparative proteomic profiling: Comparative analysis of proteomic landscapes across different human body fluids to address discrepancies between experimental studies.

Methodology:

An end-to-end sequence-based deep learning pipeline using a Convolutional Neural Network to learn features from protein sequences, followed by a Bidirectional Gated Recurrent Unit and a fully connected layer for classification, with evaluation on datasets from 12 human body fluids and application to TCGA kidney cancer genomics data.