picnic-bio.webservice

picnic-bio.webservice predicts proteins that localize to biomolecular condensates using machine learning applied to sequence- and structure-derived features.

Key Features:

• Machine Learning Algorithm: A classifier identifies proteins that localize to biomolecular condensates, irrespective of their role as drivers of formation.
• Feature Utilization: Uses amino acid patterns from protein sequences, intrinsic disorder metrics, and structural data derived from Alphafold2 models.
• Unbiased Prediction: Provides predictions across a wide range of proteins without bias toward high intrinsic disorder content.
• Experimental Validation: Achieved approximately 82% accuracy for positive predictions in cellulo experimental validation.
• Cross-Species Analysis: Evaluation across 26 species found no correlation between predicted condensate proteome content and disorder content.

Scientific Applications:

• Whole-proteome composition analysis: Enables prediction of condensate membership across entire proteomes to map condensate proteomes.
• Hypothesis generation for experiments: Prioritizes candidate proteins for in cellulo validation of condensate localization.
• Identification of non–disorder-rich components: Detects potential condensate members that lack high intrinsic disorder and might be missed by disorder-centric approaches.
• Comparative and cross-species studies: Supports comparative analysis of condensate composition across multiple species.

Methodology:

Trained a machine learning classifier on labeled condensate data using amino acid pattern features, intrinsic disorder metrics, and structural features derived from Alphafold2 models.