SETH_1

SETH_1 predicts residue disorder in proteins using embeddings from the protein language model ProtT5 and a model trained on CheZOD scores to identify intrinsically disordered regions from single-sequence input without requiring multiple sequence alignments (MSAs).

Key Features:

• Single Sequence Input: SETH_1 uses single amino acid sequences and ProtT5 embeddings, eliminating the need for multiple sequence alignments (MSAs).
• Efficient Architecture: The method employs a relatively shallow convolutional neural network to generate predictions.
• Performance and Speed: SETH_1 can produce proteome-scale predictions rapidly, reported at approximately one hour on a consumer-grade PC with an NVIDIA GeForce RTX 3060.
• Continuous Disorder Scale: The model is trained on CheZOD scores to capture continuous, residue-level variations in disorder rather than binary classifications.
• Applications in Disease Research: SETH_1 identifies intrinsically disordered regions (IDRs) that are associated with diseases, including Alzheimer’s Disease.
• Enhancing AlphaFold2 Predictions: SETH_1 can identify regions or proteins likely to yield low-quality 3D structures from AlphaFold2, enabling prioritization of computational resources.
• Evolutionary Insights: Proteome-wide disorder profiles produced by SETH_1 enable analysis of nuanced disorder variation for evolutionary studies.

Scientific Applications:

• Protein structure and function analysis: Analysis of IDRs in studies of protein structure–function relationships.
• Disease-related protein research: Investigation of proteins and regions implicated in diseases such as Alzheimer’s Disease.
• Prioritization of structure prediction: Guiding and optimizing AlphaFold2 or other large-scale 3D prediction efforts by identifying likely low-quality regions.
• Evolutionary biology: Comparative proteome analyses to study disorder variation across organisms.

Methodology:

SETH_1 generates ProtT5 embeddings from single protein sequences, processes them with a relatively shallow convolutional neural network, and is trained on CheZOD scores to produce continuous residue-level disorder predictions.