HUbiPred

HUbiPred predicts human protein ubiquitination sites using an ensemble of neural networks to improve identification of modification positions relevant to disease mechanisms.

Key Features:

• Ensemble methodology: Integrates two convolutional neural networks (CNNs) and two recurrent neural networks (RNNs), including long short-term memory (LSTM) networks.
• Feature engineering: Uses binary encoding and physicochemical properties of amino acids as input features.
• Performance metrics: Achieves area under the curve (AUC) values of 0.852 in five-fold cross-validation and 0.844 in independent tests.
• Architectural analysis: Examines contributions of convolutional layers, LSTM layers, and fully connected hidden layers to prediction performance.
• Amino acid context analysis: Enables exploration of physicochemical properties surrounding predicted ubiquitination sites.

Scientific Applications:

• Biomarker and therapeutic target discovery: Facilitates identification of ubiquitination sites relevant to diseases such as cancer.
• Post-translational modification research: Supports analysis of sequence and physicochemical determinants of ubiquitination in human proteins.

Methodology:

Employs an ensemble of two CNNs and two RNNs (LSTM) trained on binary-encoded sequences and amino acid physicochemical properties, evaluated with five-fold cross-validation and independent tests (AUCs 0.852 and 0.844), and includes analysis of convolutional, LSTM, and fully connected hidden layers.