Charting γ-secretase substrates by explainable AI.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Proteomics

Background:

• Proteases identify substrates via sequence information, a challenge when motifs are absent.
• Gamma-secretase, an intramembrane protease, is linked to Alzheimer's disease and cancer.
• Understanding protease-substrate interactions is crucial for cellular processes.

Purpose of the Study:

• To develop a novel sequence-based algorithm for identifying protease substrates without relying on traditional recognition motifs.
• To decipher the substrate signature for gamma-secretase, a key protease implicated in neurodegenerative and oncological diseases.
• To predict the full scope of human gamma-secretase substrates and uncover novel biological associations.

Main Methods:

• Development of Comparative Physicochemical Profiling (CPP), a sequence-based algorithm.
• Utilizing machine learning for predicting gamma-secretase substrate scope.
• Experimental validation of predicted substrates and identified pathways.

Main Results:

• CPP identifies interpretable physicochemical features for protease substrate recognition with single-residue resolution.
• The algorithm accurately predicts gamma-secretase substrate signatures, explaining substrate conformational changes.
• Machine learning prediction identified numerous novel gamma-secretase substrates, improving accuracy from 60% to 90% with 88% experimental validation success.
• New pathways and diseases associated with gamma-secretase activity were uncovered.

Conclusions:

• Comparative Physicochemical Profiling (CPP) effectively decodes protease substrate signatures beyond sequence motifs.
• The approach significantly advances the prediction of gamma-secretase substrates and their biological relevance.
• CPP offers a broadly applicable framework for understanding diverse molecular recognition processes.