Introduction
This special section of the IEEE/ACM Transactions on Computational Biology and Bioinformatics presents eight high-impact papers selected from the 27th International Conference on Genome Informatics (GIW2016). These works represent significant advancements in computational frameworks designed to tackle complex biological challenges, ranging from gene expression analysis to high-performance molecular simulations.
Addressing Modern Bioinformatics Challenges
The field of genome informatics currently faces several critical bottlenecks. High-dimensionality in massive gene expression datasets makes identifying meaningful patterns difficult, while inherent noise in Protein-Protein Interaction (PPI) data leads to high false-positive rates. Furthermore, predicting the functions of "dark matter" in the genome—specifically long non-coding RNAs (lncRNAs)—remains a major hurdle. Finally, the increasing complexity of molecular simulations demands better algorithmic scaling and hardware utilization.
Methodological Innovations
The selected papers introduce diverse and robust methodologies aimed at these bottlenecks:
- Subspace Weighting Co-Clustering (SWCC): Identifies the contribution of specific genes in distinguishing sample clusters using a gene subspace weight matrix.
- Reinforce: An ensemble approach for PPI network inference that stabilizes results through rank aggregation and false discovery rate (FDR) control.
- EssRank (Essentiality Ranking): Detects essential proteins by combining weighted edge clustering coefficients (WECC) with random walk algorithms.
- De Novo Ranking for Glycans: Matches MS/MS spectra with glycan structures by incorporating de novo sequencing.
- LncRDNetFlow & KATZLGO: These frameworks utilize global network flow and the KATZ measure on fused heterogeneous networks to prioritize disease-related lncRNAs and predict their functions.
- MGT-SM: Constructs cellular signal transduction networks using a multivariate Granger test-based framework.
- Parallel GROMACS: Optimizes molecular dynamics for the Tianhe-2 supercomputer through CPU and Intel Xeon Phi (MIC) collaboration.
Key Insights and Results
A recurring theme in this collection is that ensemble methods and network fusion often outperform single-algorithm approaches. For instance, the "Reinforce" method demonstrates superior stability in PPI inference, while "EssRank" shows that combining network topology with biological confidence scores is more effective than traditional centrality measures like Degree or Betweenness centrality.
The effectiveness of these methods is supported by rigorous testing. EssRank outperformed standard centrality measures and existing data-integration methods like IEW. Similarly, LncRDNetFlow and KATZLGO reported significant improvements over state-of-the-art approaches for lncRNA-disease association. In network construction, MGT-SM achieved superior results on the Yeast Synthetic Network compared to established methods. Finally, the CPU/MIC framework for GROMACS successfully enabled simultaneous processing, providing a significant speedup on the Tianhe-2 supercomputer.
Conclusion
The papers in this special section underscore the power of heterogeneous network integration and hardware-specific optimization in modern bioinformatics. As genomic datasets continue to grow in size and complexity, these methodologies provide a foundation for future research in identifying essential biological components and understanding complex cellular mechanisms. Future work will likely involve refining these network-based models as more high-quality lncRNA and protein interaction data becomes available.
Summary updated