580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can automated and precise image processing methods improve the quantification accuracy and throughput of DNA microarray data?
The processing and quantification of DNA microarray images critically impact downstream analysis reliability. Given the complexity and variability in spot morphology, grid regularity, and background noise, developing fully automated and precise image processing methods that minimize human intervention can significantly enhance accuracy, reproducibility, and throughput. This theme focuses on designing automatic gridding, spot segmentation, background correction, and ratio estimation algorithms that are robust to typical experimental deviations and spot irregularities.
Key finding: Presented a fully automatic system (UCSF Spot) that locates subarray grids and segment spots without any manual input, explicitly identifying spot pixels rather than assuming ideal spot shapes. This approach reduced user bias... Read more
Key finding: Outlined a comprehensive model of ideal microarray images and detailed real-world image deviations affecting automatic processing, emphasizing the requirements for grid layout detection, spot morphology handling, and... Read more
Key finding: Developed a fully automatic, iterative refinement method that detects and corrects rotations within microarray subarrays, increasing gridding precision without user input. By combining global and local image features along... Read more
2. What feature selection and classification strategies improve the accuracy and interpretability of gene expression analysis from DNA microarray data?
Microarray datasets are characterized by high dimensionality (thousands of genes) but often limited sample sizes, which complicates the classification and interpretation of gene expression profiles. Effective feature selection—identifying a small, informative subset of genes—combined with robust classification algorithms can improve predictive performance and reduce computational burden. This theme explores dimensionality reduction, feature ranking, and hybrid techniques to optimize gene selection and classification, including the use of evolutionary algorithms and correlation-based selection methods.
Key finding: Applied multi-layer perceptron neural networks to classify genes in yeast using log-transformed expression ratios from 79 DNA microarray experiments. The study stressed the challenges posed by imbalanced classes and noise,... Read more
Key finding: Introduced a feature selection process based on the Artificial Bee Colony (ABC) algorithm to identify informative gene subsets, followed by training artificial neural networks for classification. Validated on multiple... Read more
Key finding: Utilized Correlation-based Feature Selection (CFS) combined with Greedy Stepwise search to identify non-redundant, relevant genes, thereby reducing dimensionality and improving classification accuracy across 11 microarray... Read more
Key finding: Developed a computationally efficient filter algorithm combining F-measure and the Kolmogorov-Smirnov test to assess feature relevance and redundancy, enabling improved feature selection in high-dimensional microarray... Read more
3. How can bioinformatics tools and integrative analysis enhance the biological insight gained from DNA microarray experiments, particularly in gene regulation and functional genomics?
DNA microarrays generate vast gene expression profiles requiring sophisticated bioinformatics approaches for data integration, clustering, motif discovery, and functional annotation. This research area investigates methods to cluster co-expressed genes robustly, identify underlying transcription factor binding motifs, map enriched pathways, and connect expression with network-level biological processes. Bioinformatics frameworks and pipelines that enable multi-layered analysis advance understanding of gene regulatory mechanisms and disease-associated molecular pathways.
Key finding: Reviewed integrative computational approaches combining clustering algorithms tailored for noisy microarray data (e.g., adaptive quality-based clustering) with motif-finding techniques including Gibbs sampling to detect... Read more
Key finding: Surveyed bioinformatics tools and algorithms for key microarray data mining tasks such as differential expression detection, clustering, classification, transcription factor binding site analysis, pathway enrichment, and... Read more
Key finding: Developed an interactive framework that simultaneously visualizes multiple biclustering results alongside transcription regulatory networks and heatmaps, enabling enhanced interpretation of gene expression patterns and... Read more
Key finding: Provided a broad overview of bioinformatics evolution from sequence-based analyses to integrative functional genomics involving transcriptome profiling via DNA microarrays. Emphasized current challenges in handling... Read more