Related Experiment Videos
LiveBench-1: continuous benchmarking of protein structure prediction servers
J M Bujnicki1, A Elofsson, D Fischer
1Bioinformatics Laboratory, International Institute of Molecular and Cell Biology, 02-109 Warsaw, Poland.
Protein Science : a Publication of the Protein Society
|March 27, 2001
Summary
This study assesses protein fold-recognition servers, finding they accurately model one-third of targets. Combining server results significantly improves prediction accuracy, especially for difficult protein structures.
Area of Science:
- Computational Biology
- Structural Bioinformatics
- Protein Structure Prediction
Background:
- Protein structure prediction is crucial for understanding protein function.
- Fold-recognition servers are widely used tools for predicting protein structures.
- Evaluating the performance of these servers is essential for advancing the field.
Purpose of the Study:
- To conduct a large-scale, continuous assessment of popular protein fold-recognition server performance.
- To identify the strengths and weaknesses of different servers on various prediction tasks.
- To explore strategies for improving fold-recognition accuracy.
Main Methods:
- Evaluated six popular fold-recognition servers: PDB-Blast, FFAS, T98-lib, GenTHREADER, 3D-PSSM, and INBGU.
- Used a dataset of novel protein structures (October 1999 - April 2000) with no significant sequence similarity to existing databases.
- Classified targets into 'easy' and 'hard' categories based on prediction difficulty.
Main Results:
- Servers achieved structurally similar models for 50% of targets, but accurate sequence-structure alignments for only 33%.
- All servers performed well on 'easy' targets but struggled with 'hard' targets (40% similar models, 20% accurate alignments).
- A combined consensus approach improved correct assignments by 50%, highlighting the value of multiple servers.
Conclusions:
- Protein fold-recognition servers show variable performance, particularly on challenging targets.
- Utilizing consensus predictions from multiple servers significantly enhances accuracy.
- The LiveBench program continues this evaluation effort, inviting developer participation.