AI Decodes γ-Secretase Specificity

Researchers have developed an AI-driven method, Comparative Physicochemical Profiling (CPP), to understand how γ-secretase recognises its substrates. γ-secretase, an enzyme linked to Alzheimer's and cancer, cleaves over 150 membrane proteins, but its recognition mechanism was previously unclear. The CPP technique combines biochemistry with explainable AI (XAI) to compare physicochemical properties of known substrates against reference proteins, revealing characteristic patterns.

The study identified that γ-secretase substrates possess a specific physicochemical profile across the transmembrane domain and adjacent regions. Substrates near the cleavage site can adopt alternative conformations to their helical structure, which is essential for molecular recognition within the cell membrane. The CPP method identified previously unknown substrates, including proteins involved in immune regulation and carcinogenesis. This approach could be applied to decode sequence, structure, and function in other proteases and receptors, potentially aiding the development of therapeutic compounds with enhanced specificities.

The AI model improves prediction accuracy from 60% to 90% and achieves an 88% success rate in experimental validation. This advancement identifies pathways and diseases not previously linked to γ-secretase. The CPP algorithm deciphers a γ-secretase substrate signature with single-residue resolution, explaining conformational transitions in substrates upon γ-secretase binding.