Highly accurate protein structure prediction for the human proteome.

Area of Science:

• Structural Biology
• Computational Biology
• Genomics

Background:

• Experimental protein structures cover only 17% of human protein sequences.
• Understanding protein structure is crucial for biological insights and therapeutic development.

Purpose of the Study:

• To expand structural coverage of the human proteome using advanced machine learning.
• To provide a comprehensive dataset of predicted human protein structures for research.

Main Methods:

• Application of the AlphaFold machine learning model to predict structures for 98.5% of human proteins.
• Development and application of new metrics to assess prediction confidence and interpret structural features.

Main Results:

• Predicted structures for 98.5% of human proteins, covering 58% of residues with confident predictions.
• Identified regions with high confidence (36% of residues) and likely disordered regions.
• Demonstrated utility of predictions through case studies for generating biological hypotheses.

Conclusions:

• Large-scale, high-accuracy protein structure prediction is now feasible and accessible.
• The generated dataset significantly enhances structural coverage, enabling new research avenues.
• Freely available predictions will empower the scientific community in structure-based research.