PureseqTM

PureSeqTM predicts transmembrane (TM) topology from amino acid sequences using DeepCNF (Conditional Neural Fields) to improve TM segment identification and membrane-proteome annotation.

Key Features:

• Deep Learning Framework: PureSeqTM employs DeepCNF (Conditional Neural Fields), a hierarchical deep neural network framework that captures extensive contextual information from amino acid sequences.
• Contextual Information Integration: DeepCNF models interdependencies between adjacent topology labels more effectively than hidden Markov models or dynamic Bayesian networks, enabling richer representation of TM regions.
• Identification of Correct TM Segments: On a dataset of 39 newly released membrane proteins, PureSeqTM identified correct TM segments and boundaries in at least three cases where all existing methods failed.
• Re-evaluation of Human Proteome Annotations: Applied to the entire human proteome, PureSeqTM identified incorrect TM annotations recorded in UniProt and uncovered membrane-related proteins not manually curated in existing databases.

Scientific Applications:

• Membrane proteome annotation: Annotating the membrane proteome with high precision from amino acid sequences.
• Membrane protein structure and function prediction: Facilitating prediction of membrane protein structures and functions by providing accurate TM topology.
• Database annotation curation: Identifying potential annotation errors in protein databases such as UniProt.
• Discovery of novel membrane-related proteins: Detecting membrane-related proteins that have not been manually curated in existing resources.

Methodology:

PureSeqTM uses DeepCNF (Conditional Neural Fields), a hierarchical deep neural network, to process amino acid sequences and model interdependencies between adjacent topology labels as an alternative to hidden Markov models and dynamic Bayesian networks.