SCLpred-MEM

SCLpred-MEM predicts the subcellular localization of membrane proteins as an ab initio predictor that classifies sequences as membrane or non-membrane using an ensemble of Deep N-to-1 Convolutional Neural Networks (N1-NN) for biological annotation and analysis.

Key Features:

• Deep Learning Architecture: Uses an ensemble of Deep N-to-1 Convolutional Neural Networks (N1-NN) to perform sequence-based classification.
• Ab initio prediction: Operates without requiring homology-based transfer, relying on sequence-derived features learned by the networks.
• Dataset quality: Trained and tested on strictly homology-reduced datasets to minimize redundancy and bias.
• Performance metrics: Reports a Matthews correlation coefficient (MCC) of 0.52 on a strictly homology-reduced independent test set and MCC of 0.62 on a less strict homology-reduced dataset.
• Benchmarking: Matches or surpasses other state-of-the-art subcellular localization predictors in comparative evaluations.

Scientific Applications:

• Genomics: Assists annotation of genomic data by assigning subcellular localization to membrane protein sequences.
• Drug design: Informs target selection and localization-aware drug targeting strategies for membrane proteins.
• Theoretical and analytical bioinformatics: Provides localization predictions useful for modeling cellular processes and protein interaction contexts.

Methodology:

Ab initio prediction using an ensemble of Deep N-to-1 Convolutional Neural Networks (N1-NN) trained and evaluated on strictly homology-reduced datasets with independent test-set performance assessment.