SKADE

SKADE predicts protein solubility from protein sequence using a neural attention architecture to provide interpretable predictions and assess mutation effects.

Key Features:

• Neural network-based prediction: Predicts protein solubility using a neural network model.
• Neural attention architecture: Implements a novel attention mechanism that generates attention profiles for inspecting learned patterns and model decision-making.
• Sequence-only input: Operates using only the protein sequence as input.
• Interpretability of termini: Attention profiles highlight N- and C-termini as critical regions for solubility prediction.
• Emergent property prediction: Attention-derived signals are predictive of aggregation-prone areas involved in beta-amyloidosis and of contact density.
• In-silico mutagenesis: Identifies mutations that can increase or decrease overall protein solubility and supports large-scale in-silico mutagenesis studies.
• Benchmark performance: Demonstrates superior performance compared to existing state-of-the-art solubility prediction tools.

Scientific Applications:

• Biotechnological protein production: Inform design and optimization of active proteins and antibodies for production.
• Disease research: Study mechanisms of aggregation-related diseases such as beta-amyloidosis.
• Protein engineering: Guide mutational strategies to improve protein solubility, stability, and functionality.
• Large-scale mutagenesis studies: Enable high-throughput in-silico screening of variant effects on solubility.
• Aggregation and structural studies: Link sequence-derived attention signals to aggregation-prone regions and contact density analyses.

Methodology:

Uses a neural network with a neural attention architecture that takes protein sequence as input to predict solubility, produce attention profiles, and evaluate effects of in-silico mutations.