Related Experiment Videos
Think globally and solve locally: secondary memory-based network learning for automated multi-species function
Marco Mesiti1, Matteo Re1, Giorgio Valentini1
1AnacletoLab - Department of Computer Science, University of Milano, Via Comelico 39/41, 20135 Milano, Italy.
Gigascience
|May 21, 2014
Summary
This study introduces a scalable framework for multi-species protein function prediction, overcoming computational limits with innovative secondary memory technologies. This approach enables analysis of large biological networks on standard computers, advancing automated function prediction for poorly annotated species.
Area of Science:
- Bioinformatics
- Computational Biology
- Systems Biology
Background:
- Automated function prediction (AFP) is hindered by limited experimental data and functional annotations, restricting its use in model organisms.
- Existing network-based algorithms struggle with scalability and memory requirements for large, multi-species biological networks.
- Poorly annotated species pose a significant challenge for current AFP methods.
Purpose of the Study:
- To develop a scalable framework for network-based learning of multi-species protein functions.
- To overcome the computational limitations of existing AFP algorithms when applied to large biological networks.
- To enable effective function prediction for poorly annotated species by leveraging multi-species network analysis.
Main Methods:
- Developed a novel framework combining local, vertex-centric implementations of network-based algorithms with global network topology analysis.
- Utilized secondary memory-based technologies to overcome primary memory limitations, enabling efficient processing of large datasets.
- Applied the framework to analyze a multi-species network of over 300 bacterial species and a network of over 200,000 proteins from 13 eukaryotic species.
Main Results:
- Successfully implemented a scalable approach for multi-species protein function prediction.
- Demonstrated the feasibility of analyzing large biological networks (hundreds of thousands of proteins) using secondary memory-based techniques.
- Achieved multi-species function prediction on a scale previously unattainable with standard computational resources.
Conclusions:
- The proposed algorithmic and technological approaches enable the analysis of large multi-species networks on ordinary computers.
- This framework has the potential to facilitate the analysis of entire proteomes (e.g., SwissProt) on well-equipped standalone machines.
- The study represents a significant advancement in applying secondary-memory based network analysis to multi-species function prediction.