AF2Complex: Benchmark Data and Input Features for E. coli Protein Interaction Modeling

Benchmark data sets and pre-generated input features for the AF2Complex deep learning method, focusing on protein complex prediction. The data includes benchmark sets CP17, Dimer1193, and Oligomer562, as well as features for 4,429 proteins from the E. coli proteome. This dataset was created by Mu Gao of the Georgia Institute of Technology to support the work 'Predicting direct physical interactions in multimeric proteins with deep learning'.

Use Cases

• Benchmarking protein complex prediction algorithms based on the provided CP17, Dimer1193, and Oligomer562 test sets.
• Probing protein-protein interactions within the E. coli proteome using the pre-generated input features for 4,429 proteins.
• Training or fine-tuning deep learning models for structural biology using the provided multiple sequence alignments and structural models.
• Validating predicted physical interactions against the provided lists of positive and negative protein-protein interactions from E. coli.

Strengths

• Includes three distinct benchmark datasets (CP17, Dimer1193, Oligomer562) for method evaluation.
• Provides pre-computed input features for the entire E. coli proteome, covering 4,429 proteins.
• Contains experimental structures and computational models, such as those for the E. coli Ccm I system.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The total decompressed size is very large (~890 GB), which may pose storage and handling challenges.

Provenance

Source

Georgia Institute of Technology

Collection Method

Computational generation and curation for the AF2Complex method, as described in the associated research paper.