Celebrating our 20th digital anniversary

20 Years of Highly Cited Discoveries Published in PNAS

Organizing complex genetic data

Researchers report that clustering gene-expression data derived from DNA microarray hybridization into groups of genes with similar patterns organizes them by function, and suggest that displaying the clustered data using a color matrix allows researchers to assimilate and explore complex genomic data in a natural and intuitive manner.

Eisen MB, Spellman PT, Brown PO, Botstein D (1998) Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 95(25):14863–14868.

Baseline brain activity

PET scans of 49 healthy people, aged 19–84 years, helped identify a baseline activity state for the adult human brain in people at rest and with their eyes closed, as measured by the ratio of oxygen that the brain consumes to oxygen delivered to the brain, suggesting that a baseline default mode of brain function exists and is attenuated in specific brain regions during certain goal-directed behaviors.

Raichle ME, et al. (2001) A default mode of brain function. Proc Natl Acad Sci USA 98(2):676–682.

Genetics of breast cancer

According to an analysis of gene-expression patterns of 78 human breast carcinomas and 7 noncancerous samples, estrogen receptor-positive tumors can be divided into two subgroups in accordance with two sets of genes that are associated with tumor properties and patient outcomes; survival analyses of a subcohort of patients with locally advanced breast cancer and similar treatments found that patient outcomes differed significantly among tumors with different gene-expression patterns.

Sørlie T, et al. (2001) Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 98(19):10869–10874.

Genomic statistics

Researchers report a statistical method that estimates the false-discovery rate, or the rate that significant features are truly null, of each tested feature of genome-wide studies using a parameter called the q value; the approach might help avoid an excess of false-positive results and allows a liberal criterion for highlighting features for future investigation.

Storey JD, Tibshirani R (2003) Statistical significance for genomewide studies. Proc Natl Acad Sci USA100(16):9440–9445.