Guest Editorial for Special Section on BIOKDD2013

Gaurav Pandey, Huzefa RangwalaIEEE/ACM 2014 · DOI · OpenAlex

Abstract

The four articles in this special section were presented at the 2013 International Workshop on Data Mining in Bioinformatics (BIOKDD) which was held in conjunction with the ACM Conference on Knowledge Discovery and Data Mining (KDD) in Chicago, IL.

Introduction

This special section of the IEEE/ACM Transactions on Computational Biology and Bioinformatics showcases four standout papers from the 12th International Workshop on Data Mining in Bioinformatics (BIOKDD 2013). Guest editors Gaurav Pandey and Huzefa Rangwala curated these works to highlight advanced data mining techniques—ranging from probabilistic logic to suffix trees—designed to navigate the "big data" explosion in modern biomedicine.

Gaurav Pandey

The Big Data Challenge in Biomedicine

Biological sciences are currently awash in data from high-throughput sequencing, imaging, and proteomics. The primary bottleneck has shifted from data generation to data integration and hypothesis testing. The "search space" for valid biological discoveries is simply too vast for traditional methods. The featured research provides scalable solutions for drug repositioning, genomic mapping, time-series gene expression analysis, and 3D protein structure discovery.

Methodological Innovations

The editorial summarizes four distinct methodological approaches tailored to specific biological problems:

1. Drug Discovery via Probabilistic Soft Logic (PSL)

To address the difficulty of identifying new uses for existing drugs (repositioning), researchers utilized PSL for drug-target interaction prediction. By representing relationships as a bipartite graph and applying PSL-derived rules, the system can infer new links between drugs and targets. This approach suggests that interactions are better modeled with "shades of gray" and rule-based reasoning rather than simple binary links.

2. Genomic Mapping with Divide-and-Conquer Clustering

Radiation hybrid mapping (RHM) experiments are often computationally expensive. This research applies divide-and-conquer clustering to break down the problem into smaller, manageable clusters. This approach produced "better framework maps" than state-of-the-art tools while significantly reducing computational complexity.

3. Time-Series Gene Expression and Suffix Trees

Identifying "time-lagged" patterns in gene activity—where one gene's expression influences another after a delay—is typically an NP-hard problem. A novel algorithm called LateBiclustering utilizes suffix tree-based heuristics to identify these patterns in linear time. The algorithm successfully identified groups of genes that were not just mathematically similar but functionally related.

4. Protein Function through Spatial Proximity Itemset Mining

Identifying functional sites within complex 3D protein structures is a major challenge. Researchers defined a "support function" based on the physical distance between amino acids rather than their sequence order. This spatial itemset mining successfully identified known biologically important entities, such as enzyme active sites, demonstrating that geometry often supersedes linear sequence in defining function.

Key Insights and Results

The featured papers collectively demonstrate that:

  • Scalability is Essential: As genomic datasets grow, "divide-and-conquer" and heuristic approaches are necessary to keep computational costs manageable.
  • Heuristics for Complexity: The LateBiclustering paper proves that with the right heuristic (suffix trees), biologically significant patterns can be found much faster than previously thought possible.
  • Structural Context: In protein analysis, functional sites are defined by how amino acids fold together in 3D space, emphasizing the importance of spatial data over raw sequence strings.

Conclusion and Future Work

While these papers represent significant strides, the editors emphasize the ongoing need for more "integrative" data analysis that combines diverse data types (e.g., combining 3D structure with time-series expression). Furthermore, as the scale of biomedical "big data" continues to increase, the development of scalable methods and rigorous literature-based and experimental validation remains paramount for the field.

Summary updated

Abstract

The four articles in this special section were presented at the 2013 International Workshop on Data Mining in Bioinformatics (BIOKDD) which was held in conjunction with the ACM Conference on Knowledge Discovery and Data Mining (KDD) in Chicago, IL.
BibTeX
@inproceedings{pandey2014guest,
  author = {Gaurav Pandey and Huzefa Rangwala},
  title = {{Guest Editorial for Special Section on BIOKDD2013}},
  booktitle = {IEEE/ACM},
  year = {2014},
  doi = {10.1109/tcbb.2014.2348731},
}

Loading related content...

Figure