Maternal diabetes and obesity influence the fetal epigenome in a largely Hispanic population.

BACKGROUND: Obesity and diabetes mellitus are directly implicated in many adverse health consequences in adults as well as in the offspring of obese and diabetic mothers. Hispanic Americans are particularly at risk for obesity, diabetes, and end-stage renal disease. Maternal obesity and/or diabetes through prenatal programming may alter the fetal epigenome increasing the risk of metabolic disease in their offspring. The aims of this study were to determine if maternal obesity or diabetes mellitus during pregnancy results in a change in infant methylation of CpG islands adjacent to targeted genes specific for obesity or diabetes disease pathways in a largely Hispanic population. METHODS: Methylation levels in the cord blood of 69 newborns were determined using the Illumina Infinium MethylationEPIC BeadChip. Over 850,000 different probe sites were analyzed to determine whether maternal obesity and/or diabetes mellitus directly attributed to differential methylation; epigenome-wide and regional analyses were performed for significant CpG sites. RESULTS: Following quality control, agranular leukocyte samples from 69 newborns (23 normal term (NT), 14 diabetes (DM), 23 obese (OB), 9 DM/OB) were analyzed for over 850,000 different probe sites. Contrasts between the NT, DM, OB, and DM/OB were considered. After correction for multiple testing, 15 CpGs showed differential methylation from the NT, associated with 10 differentially methylated genes between the diabetic and non-diabetic subgroups, CCDC110, KALRN, PAG1, GNRH1, SLC2A9, CSRP2BP, HIVEP1, RALGDS, DHX37, and SCNN1D. The effects of diabetes were partly mediated by the altered methylation of HOOK2, LCE3C, and TMEM63B. The effects of obesity were partly mediated by the differential methylation of LTF and DUSP22. CONCLUSIONS: The presented data highlights the associated altered methylation patterns potentially mediated by maternal diabetes and/or obesity. Larger studies are warranted to investigate the role of both the identified differentially methylated loci and the effects on newborn body composition and future health risk factors for metabolic disease. Additional future consideration should be targeted to the role of Hispanic inheritance. Potential future targeting of transgenerational propagation and developmental programming may reduce population obesity and diabetes risk.

Regulatory complexity revealed by integrated cytological and RNA-seq analyses of meiotic substages in mouse spermatocytes.

BACKGROUND: The continuous and non-synchronous nature of postnatal male germ-cell development has impeded stage-specific resolution of molecular events of mammalian meiotic prophase in the testis. Here the juvenile onset of spermatogenesis in mice is analyzed by combining cytological and transcriptomic data in a novel computational analysis that allows decomposition of the transcriptional programs of spermatogonia and meiotic prophase substages. RESULTS: Germ cells from testes of individual mice were obtained at two-day intervals from 8 to 18 days post-partum (dpp), prepared as surface-spread chromatin and immunolabeled for meiotic stage-specific protein markers (STRA8, SYCP3, phosphorylated H2AFX, and HISTH1T). Eight stages were discriminated cytologically by combinatorial antibody labeling, and RNA-seq was performed on the same samples. Independent principal component analyses of cytological and transcriptomic data yielded similar patterns for both data types, providing strong evidence for substage-specific gene expression signatures. A novel permutation-based maximum covariance analysis (PMCA) was developed to map co-expressed transcripts to one or more of the eight meiotic prophase substages, thereby linking distinct molecular programs to cytologically defined cell states. Expression of meiosis-specific genes is not substage-limited, suggesting regulation of substage transitions at other levels. CONCLUSIONS: This integrated analysis provides a general method for resolving complex cell populations. Here it revealed not only features of meiotic substage-specific gene expression, but also a network of substage-specific transcription factors and relationships to potential target genes.

Unraveling the message: insights into comparative genomics of the naked mole-rat.

Animals have evolved to survive, and even thrive, in different environments. Genetic adaptations may have indirectly created phenotypes that also resulted in a longer lifespan. One example of this phenomenon is the preternaturally long-lived naked mole-rat. This strictly subterranean rodent tolerates hypoxia, hypercapnia, and soil-based toxins. Naked mole-rats also exhibit pronounced resistance to cancer and an attenuated decline of many physiological characteristics that often decline as mammals age. Elucidating mechanisms that give rise to their unique phenotypes will lead to better understanding of subterranean ecophysiology and biology of aging. Comparative genomics could be a useful tool in this regard. Since the publication of a naked mole-rat genome assembly in 2011, analyses of genomic and transcriptomic data have enabled a clearer understanding of mole-rat evolutionary history and suggested molecular pathways (e.g., NRF2-signaling activation and DNA damage repair mechanisms) that may explain the extraordinarily longevity and unique health traits of this species. However, careful scrutiny and re-analysis suggest that some identified features result from incorrect or imprecise annotation and assembly of the naked mole-rat genome: in addition, some of these conclusions (e.g., genes involved in cancer resistance and hairlessness) are rejected when the analysis includes additional, more closely related species. We describe how the combination of better study design, improved genomic sequencing techniques, and new bioinformatic and data analytical tools will improve comparative genomics and ultimately bridge the gap between traditional model and nonmodel organisms.

Homology cluster differential expression analysis for interspecies mRNA-Seq experiments.

There is an increasing demand for exploration of the transcriptomes of multiple species with extraordinary traits such as the naked-mole rat (NMR). The NMR is remarkable because of its longevity and resistance to developing cancer. It is of scientific interest to understand the molecular mechanisms that impart these traits, and RNA-sequencing experiments with comparator species can correlate transcriptome dynamics with these phenotypes. Comparing transcriptome differences requires a homology mapping of each transcript in one species to transcript(s) within the other. Such mappings are necessary, especially if one species does not have well-annotated genome available. Current approaches for this type of analysis typically identify the best match for each transcript, but the best match analysis ignores the inherent risks of mismatch when there are multiple candidate transcripts with similar homology scores. We present a method that treats the set of homologs from a novel species as a cluster corresponding to a single gene in the reference species, and we compare the cluster-based approach to a conventional best-match analysis in both simulated data and a case study with NMR and mouse tissues. We demonstrate that the cluster-based approach has superior power to detect differential expression.

Whole-genome sequencing identifies EN1 as a determinant of bone density and fracture.

The extent to which low-frequency (minor allele frequency (MAF) between 1-5%) and rare (MAF ≤ 1%) variants contribute to complex traits and disease in the general population is mainly unknown. Bone mineral density (BMD) is highly heritable, a major predictor of osteoporotic fractures, and has been previously associated with common genetic variants, as well as rare, population-specific, coding variants. Here we identify novel non-coding genetic variants with large effects on BMD (ntotal = 53,236) and fracture (ntotal = 508,253) in individuals of European ancestry from the general population. Associations for BMD were derived from whole-genome sequencing (n = 2,882 from UK10K (ref. 10); a population-based genome sequencing consortium), whole-exome sequencing (n = 3,549), deep imputation of genotyped samples using a combined UK10K/1000 Genomes reference panel (n = 26,534), and de novo replication genotyping (n = 20,271). We identified a low-frequency non-coding variant near a novel locus, EN1, with an effect size fourfold larger than the mean of previously reported common variants for lumbar spine BMD (rs11692564(T), MAF = 1.6%, replication effect size = +0.20 s.d., Pmeta = 2 × 10(-14)), which was also associated with a decreased risk of fracture (odds ratio = 0.85; P = 2 × 10(-11); ncases = 98,742 and ncontrols = 409,511). Using an En1(cre/flox) mouse model, we observed that conditional loss of En1 results in low bone mass, probably as a consequence of high bone turnover. We also identified a novel low-frequency non-coding variant with large effects on BMD near WNT16 (rs148771817(T), MAF = 1.2%, replication effect size = +0.41 s.d., Pmeta = 1 × 10(-11)). In general, there was an excess of association signals arising from deleterious coding and conserved non-coding variants. These findings provide evidence that low-frequency non-coding variants have large effects on BMD and fracture, thereby providing rationale for whole-genome sequencing and improved imputation reference panels to study the genetic architecture of complex traits and disease in the general population.

Affinity-seq detects genome-wide PRDM9 binding sites and reveals the impact of prior chromatin modifications on mammalian recombination hotspot usage.

BACKGROUND: Genetic recombination plays an important role in evolution, facilitating the creation of new, favorable combinations of alleles and the removal of deleterious mutations by unlinking them from surrounding sequences. In most mammals, the placement of genetic crossovers is determined by the binding of PRDM9, a highly polymorphic protein with a long zinc finger array, to its cognate binding sites. It is one of over 800 genes encoding proteins with zinc finger domains in the human genome. RESULTS: We report a novel technique, Affinity-seq, that for the first time identifies both the genome-wide binding sites of DNA-binding proteins and quantitates their relative affinities. We have applied this in vitro technique to PRDM9, the zinc-finger protein that activates genetic recombination, obtaining new information on the regulation of hotspots, whose locations and activities determine the recombination landscape. We identified 31,770 binding sites in the mouse genome for the PRDM9(Dom2) variant. Comparing these results with hotspot usage in vivo, we find that less than half of potential PRDM9 binding sites are utilized in vivo. We show that hotspot usage is increased in actively transcribed genes and decreased in genomic regions containing H3K9me2/3 histone marks or bound to the nuclear lamina. CONCLUSIONS: These results show that a major factor determining whether a binding site will become an active hotspot and what its activity will be are constraints imposed by prior chromatin modifications on the ability of PRDM9 to bind to DNA in vivo. These constraints lead to the presence of long genomic regions depleted of recombination.

RNA-Seq alignment to individualized genomes improves transcript abundance estimates in multiparent populations.

Massively parallel RNA sequencing (RNA-seq) has yielded a wealth of new insights into transcriptional regulation. A first step in the analysis of RNA-seq data is the alignment of short sequence reads to a common reference genome or transcriptome. Genetic variants that distinguish individual genomes from the reference sequence can cause reads to be misaligned, resulting in biased estimates of transcript abundance. Fine-tuning of read alignment algorithms does not correct this problem. We have developed Seqnature software to construct individualized diploid genomes and transcriptomes for multiparent populations and have implemented a complete analysis pipeline that incorporates other existing software tools. We demonstrate in simulated and real data sets that alignment to individualized transcriptomes increases read mapping accuracy, improves estimation of transcript abundance, and enables the direct estimation of allele-specific expression. Moreover, when applied to expression QTL mapping we find that our individualized alignment strategy corrects false-positive linkage signals and unmasks hidden associations. We recommend the use of individualized diploid genomes over reference sequence alignment for all applications of high-throughput sequencing technology in genetically diverse populations.

Phenotypic dissection of bone mineral density reveals skeletal site specificity and facilitates the identification of novel loci in the genetic regulation of bone mass attainment.

Heritability of bone mineral density (BMD) varies across skeletal sites, reflecting different relative contributions of genetic and environmental influences. To quantify the degree to which common genetic variants tag and environmental factors influence BMD, at different sites, we estimated the genetic (rg) and residual (re) correlations between BMD measured at the upper limbs (UL-BMD), lower limbs (LL-BMD) and skull (SK-BMD), using total-body DXA scans of ∼ 4,890 participants recruited by the Avon Longitudinal Study of Parents and their Children (ALSPAC). Point estimates of rg indicated that appendicular sites have a greater proportion of shared genetic architecture (LL-/UL-BMD rg = 0.78) between them, than with the skull (UL-/SK-BMD rg = 0.58 and LL-/SK-BMD rg = 0.43). Likewise, the residual correlation between BMD at appendicular sites (r(e) = 0.55) was higher than the residual correlation between SK-BMD and BMD at appendicular sites (r(e) = 0.20-0.24). To explore the basis for the observed differences in rg and re, genome-wide association meta-analyses were performed (n ∼ 9,395), combining data from ALSPAC and the Generation R Study identifying 15 independent signals from 13 loci associated at genome-wide significant level across different skeletal regions. Results suggested that previously identified BMD-associated variants may exert site-specific effects (i.e. differ in the strength of their association and magnitude of effect across different skeletal sites). In particular, variants at CPED1 exerted a larger influence on SK-BMD and UL-BMD when compared to LL-BMD (P = 2.01 × 10(-37)), whilst variants at WNT16 influenced UL-BMD to a greater degree when compared to SK- and LL-BMD (P = 2.31 × 10(-14)). In addition, we report a novel association between RIN3 (previously associated with Paget's disease) and LL-BMD (rs754388: ß = 0.13, SE = 0.02, P = 1.4 × 10(-10)). Our results suggest that BMD at different skeletal sites is under a mixture of shared and specific genetic and environmental influences. Allowing for these differences by performing genome-wide association at different skeletal sites may help uncover new genetic influences on BMD.

Novel insights into embryonic stem cell self-renewal revealed through comparative human and mouse systems biology networks.

Embryonic stem cells (ESCs), characterized by their ability to both self-renew and differentiate into multiple cell lineages, are a powerful model for biomedical research and developmental biology. Human and mouse ESCs share many features, yet have distinctive aspects, including fundamental differences in the signaling pathways and cell cycle controls that support self-renewal. Here, we explore the molecular basis of human ESC self-renewal using Bayesian network machine learning to integrate cell-type-specific, high-throughput data for gene function discovery. We integrated high-throughput ESC data from 83 human studies (~1.8 million data points collected under 1,100 conditions) and 62 mouse studies (~2.4 million data points collected under 1,085 conditions) into separate human and mouse predictive networks focused on ESC self-renewal to analyze shared and distinct functional relationships among protein-coding gene orthologs. Computational evaluations show that these networks are highly accurate, literature validation confirms their biological relevance, and reverse transcriptase polymerase chain reaction (RT-PCR) validation supports our predictions. Our results reflect the importance of key regulatory genes known to be strongly associated with self-renewal and pluripotency in both species (e.g., POU5F1, SOX2, and NANOG), identify metabolic differences between species (e.g., threonine metabolism), clarify differences between human and mouse ESC developmental signaling pathways (e.g., leukemia inhibitory factor (LIF)-activated JAK/STAT in mouse; NODAL/ACTIVIN-A-activated fibroblast growth factor in human), and reveal many novel genes and pathways predicted to be functionally associated with self-renewal in each species. These interactive networks are available online at www.StemSight.org for stem cell researchers to develop new hypotheses, discover potential mechanisms involving sparsely annotated genes, and prioritize genes of interest for experimental validation.

Cell-type-specific predictive network yields novel insights into mouse embryonic stem cell self-renewal and cell fate.

Self-renewal, the ability of a stem cell to divide repeatedly while maintaining an undifferentiated state, is a defining characteristic of all stem cells. Here, we clarify the molecular foundations of mouse embryonic stem cell (mESC) self-renewal by applying a proven Bayesian network machine learning approach to integrate high-throughput data for protein function discovery. By focusing on a single stem-cell system, at a specific developmental stage, within the context of well-defined biological processes known to be active in that cell type, we produce a consensus predictive network that reflects biological reality more closely than those made by prior efforts using more generalized, context-independent methods. In addition, we show how machine learning efforts may be misled if the tissue specific role of mammalian proteins is not defined in the training set and circumscribed in the evidential data. For this study, we assembled an extensive compendium of mESC data: ∼2.2 million data points, collected from 60 different studies, under 992 conditions. We then integrated these data into a consensus mESC functional relationship network focused on biological processes associated with embryonic stem cell self-renewal and cell fate determination. Computational evaluations, literature validation, and analyses of predicted functional linkages show that our results are highly accurate and biologically relevant. Our mESC network predicts many novel players involved in self-renewal and serves as the foundation for future pluripotent stem cell studies. This network can be used by stem cell researchers (at http://StemSight.org) to explore hypotheses about gene function in the context of self-renewal and to prioritize genes of interest for experimental validation.

Tissue-specific functional networks for prioritizing phenotype and disease genes.

Integrated analyses of functional genomics data have enormous potential for identifying phenotype-associated genes. Tissue-specificity is an important aspect of many genetic diseases, reflecting the potentially different roles of proteins and pathways in diverse cell lineages. Accounting for tissue specificity in global integration of functional genomics data is challenging, as "functionality" and "functional relationships" are often not resolved for specific tissue types. We address this challenge by generating tissue-specific functional networks, which can effectively represent the diversity of protein function for more accurate identification of phenotype-associated genes in the laboratory mouse. Specifically, we created 107 tissue-specific functional relationship networks through integration of genomic data utilizing knowledge of tissue-specific gene expression patterns. Cross-network comparison revealed significantly changed genes enriched for functions related to specific tissue development. We then utilized these tissue-specific networks to predict genes associated with different phenotypes. Our results demonstrate that prediction performance is significantly improved through using the tissue-specific networks as compared to the global functional network. We used a testis-specific functional relationship network to predict genes associated with male fertility and spermatogenesis phenotypes, and experimentally confirmed one top prediction, Mbyl1. We then focused on a less-common genetic disease, ataxia, and identified candidates uniquely predicted by the cerebellum network, which are supported by both literature and experimental evidence. Our systems-level, tissue-specific scheme advances over traditional global integration and analyses and establishes a prototype to address the tissue-specific effects of genetic perturbations, diseases and drugs.

Canonical A-to-I and C-to-U RNA editing is enriched at 3'UTRs and microRNA target sites in multiple mouse tissues.

RNA editing is a process that modifies RNA nucleotides and changes the efficiency and fidelity of the central dogma. Enzymes that catalyze RNA editing are required for life, and defects in RNA editing are associated with many diseases. Recent advances in sequencing have enabled the genome-wide identification of RNA editing sites in mammalian transcriptomes. Here, we demonstrate that canonical RNA editing (A-to-I and C-to-U) occurs in liver, white adipose, and bone tissues of the laboratory mouse, and we show that apparent non-canonical editing (all other possible base substitutions) is an artifact of current high-throughput sequencing technology. Further, we report that high-confidence canonical RNA editing sites can cause non-synonymous amino acid changes and are significantly enriched in 3' UTRs, specifically at microRNA target sites, suggesting both regulatory and functional consequences for RNA editing.

Genome-wide analysis of N1ICD/RBPJ targets in vivo reveals direct transcriptional regulation of Wnt, SHH, and hippo pathway effectors by Notch1.

The Notch pathway plays a pivotal role in regulating cell fate decisions in many stem cell systems. However, the full repertoire of Notch target genes in vivo and the mechanisms through which this pathway activity is integrated with other signaling pathways are largely unknown. Here, we report a transgenic mouse in which the activation of the Notch pathway massively expands the neural stem cell (NSC) pool in a cell context-dependent manner. Using this in vivo system, we identify direct targets of RBPJ/N1ICD in cortical NSCs at a genome-wide level through combined ChIP-Seq and transcriptome analyses. Through a highly conservative analysis of these datasets, we identified 98 genes that are directly regulated by N1ICD/RPBJ in vivo. These include many transcription factors that are known to be critical for NSC self-renewal (Sox2, Pax6, Tlx, and Id4) and the transcriptional effectors of the Wnt, SHH, and Hippo pathways, TCF4, Gli2, Gli3, Yap1, and Tead2. Since little is known about the function of the Hippo-Yap pathway in NSCs, we analyzed Yap1 expression and function in NSCs. We show that Yap1 expression is restricted to the stem cell compartment in the developing forebrain and that its expression is sufficient to rescue Notch pathway inhibition in NSC self-renewal assays. Together, results of this study reveal a previously underappreciated complexity and breadth of Notch1 targets in vivo and show direct interaction between Notch and Hippo-Yap pathways in NSCs.

The need for mouse models in osteoporosis genetics research.

Osteoporosis, the progressive loss of bone mass resulting in fragility fractures, affects ∼75 million people in the United States, Europe and Japan. Bone mineral density (BMD) correlates with fracture risk and is widely used in clinical settings to predict fracture. Numerous studies have demonstrated that peak bone mass is highly heritable and consequently a number of genome-wide association studies (GWASs) have been conducted to identify the genes that regulate BMD. Traditional intercross mapping in the mouse has met with limited successes in the field of skeletal biology. With the advent of human GWAS, questions have arisen about the continued need for mouse models in genetics research. However, significant advances have been made in the field of mouse genetics, including new genetics resource populations and loci mapping techniques, which enable gene-level mapping resolution. In this review, we discuss the need for mouse models to help understand the skeletal biology underlying novel human GWAS findings, how loci discovered in the mouse can be used to complement GWAS analysis and highlight the recent advances made in the field of skeletal biology from the use of these new and developing resources. We conclude this paper with a discussion of the need for systems-level approaches in the skeletal biology field, with an emphasis on the need for pathway and network analyses.

Molecular clustering identifies complement and endothelin induction as early events in a mouse model of glaucoma.

Glaucoma is one of the most common neurodegenerative diseases. Despite this, the earliest stages of this complex disease are still unclear. This study was specifically designed to identify early stages of glaucoma in DBA/2J mice. To do this, we used genome-wide expression profiling of optic nerve head and retina and a series of computational methods. Eyes with no detectable glaucoma by conventional assays were grouped into molecularly defined stages of disease using unbiased hierarchical clustering. These stages represent a temporally ordered sequence of glaucoma states. We then determined networks and biological processes that were altered at these early stages. Early-stage expression changes included upregulation of both the complement cascade and the endothelin system, and so we tested the therapeutic value of separately inhibiting them. Mice with a mutation in complement component 1a (C1qa) were protected from glaucoma. Similarly, inhibition of the endothelin system with bosentan, an endothelin receptor antagonist, was strongly protective against glaucomatous damage. Since endothelin 2 is potently vasoconstrictive and was produced by microglia/macrophages, our data provide what we believe to be a novel link between these cell types and vascular dysfunction in glaucoma. Targeting early molecular events, such as complement and endothelin induction, may provide effective new treatments for human glaucoma.

Quantitative analysis of fitness and genetic interactions in yeast on a genome scale.

Global quantitative analysis of genetic interactions is a powerful approach for deciphering the roles of genes and mapping functional relationships among pathways. Using colony size as a proxy for fitness, we developed a method for measuring fitness-based genetic interactions from high-density arrays of yeast double mutants generated by synthetic genetic array (SGA) analysis. We identified several experimental sources of systematic variation and developed normalization strategies to obtain accurate single- and double-mutant fitness measurements, which rival the accuracy of other high-resolution studies. We applied the SGA score to examine the relationship between physical and genetic interaction networks, and we found that positive genetic interactions connect across functionally distinct protein complexes revealing a network of genetic suppression among loss-of-function alleles.

Functional genomics complements quantitative genetics in identifying disease-gene associations.

An ultimate goal of genetic research is to understand the connection between genotype and phenotype in order to improve the diagnosis and treatment of diseases. The quantitative genetics field has developed a suite of statistical methods to associate genetic loci with diseases and phenotypes, including quantitative trait loci (QTL) linkage mapping and genome-wide association studies (GWAS). However, each of these approaches have technical and biological shortcomings. For example, the amount of heritable variation explained by GWAS is often surprisingly small and the resolution of many QTL linkage mapping studies is poor. The predictive power and interpretation of QTL and GWAS results are consequently limited. In this study, we propose a complementary approach to quantitative genetics by interrogating the vast amount of high-throughput genomic data in model organisms to functionally associate genes with phenotypes and diseases. Our algorithm combines the genome-wide functional relationship network for the laboratory mouse and a state-of-the-art machine learning method. We demonstrate the superior accuracy of this algorithm through predicting genes associated with each of 1157 diverse phenotype ontology terms. Comparison between our prediction results and a meta-analysis of quantitative genetic studies reveals both overlapping candidates and distinct, accurate predictions uniquely identified by our approach. Focusing on bone mineral density (BMD), a phenotype related to osteoporotic fracture, we experimentally validated two of our novel predictions (not observed in any previous GWAS/QTL studies) and found significant bone density defects for both Timp2 and Abcg8 deficient mice. Our results suggest that the integration of functional genomics data into networks, which itself is informative of protein function and interactions, can successfully be utilized as a complementary approach to quantitative genetics to predict disease risks. All supplementary material is available at http://cbfg.jax.org/phenotype.

Visualization of omics data for systems biology.

High-throughput studies of biological systems are rapidly accumulating a wealth of 'omics'-scale data. Visualization is a key aspect of both the analysis and understanding of these data, and users now have many visualization methods and tools to choose from. The challenge is to create clear, meaningful and integrated visualizations that give biological insight, without being overwhelmed by the intrinsic complexity of the data. In this review, we discuss how visualization tools are being used to help interpret protein interaction, gene expression and metabolic profile data, and we highlight emerging new directions.

Computational analysis of the yeast proteome: understanding and exploiting functional specificity in genomic data.

Modern experimental techniques have produced a wealth of high-throughput data that has enabled the ongoing genomic revolution. As the field continues to integrate experimental and computational analyzes of this data, it is essential that performance evaluations of high-throughput results be carried out in a consistent and biologically informative manner. Here, we present an overview of evaluation techniques for high-throughput experimental data and computational methods, and we discuss a number of potential pitfalls in this process. These primarily involve the biological diversity of genomic data, which can be masked or misrepresented in overly simplified global evaluations. We describe systems for preserving information about biological context during dataset evaluation, which can help to ensure that multiple different evaluations are more directly comparable. This biological variety in high-throughput data can also be taken advantage of computationally through data integration and process specificity to produce richer systems-level predictions of cellular function. An awareness of these considerations can greatly improve the evaluation and analysis of any high-throughput experimental dataset.

The impact of incomplete knowledge on evaluation: an experimental benchmark for protein function prediction.

MOTIVATION: Rapidly expanding repositories of highly informative genomic data have generated increasing interest in methods for protein function prediction and inference of biological networks. The successful application of supervised machine learning to these tasks requires a gold standard for protein function: a trusted set of correct examples, which can be used to assess performance through cross-validation or other statistical approaches. Since gene annotation is incomplete for even the best studied model organisms, the biological reliability of such evaluations may be called into question. RESULTS: We address this concern by constructing and analyzing an experimentally based gold standard through comprehensive validation of protein function predictions for mitochondrion biogenesis in Saccharomyces cerevisiae. Specifically, we determine that (i) current machine learning approaches are able to generalize and predict novel biology from an incomplete gold standard and (ii) incomplete functional annotations adversely affect the evaluation of machine learning performance. While computational approaches performed better than predicted in the face of incomplete data, relative comparison of competing approaches-even those employing the same training data-is problematic with a sparse gold standard. Incomplete knowledge causes individual methods' performances to be differentially underestimated, resulting in misleading performance evaluations. We provide a benchmark gold standard for yeast mitochondria to complement current databases and an analysis of our experimental results in the hopes of mitigating these effects in future comparative evaluations. AVAILABILITY: The mitochondrial benchmark gold standard, as well as experimental results and additional data, is available at http://function.princeton.edu/mitochondria.