A genome sequencing system for universal newborn screening, diagnosis, and precision medicine for severe genetic diseases

Newborn screening (NBS) dramatically improves outcomes in severe childhood disorders by treatment before symptom onset. In many genetic diseases, however, outcomes remain poor because NBS has lagged behind drug development. Rapid whole-genome sequencing (rWGS) is attractive for comprehensive NBS because it concomitantly examines almost all genetic diseases and is gaining acceptance for genetic disease diagnosis in ill newborns. We describe prototypic methods for scalable, parentally consented, feedback-informed NBS and diagnosis of genetic diseases by rWGS and virtual, acute management guidance (NBS-rWGS). Using established criteria and the Delphi method, we reviewed 457 genetic diseases for NBS-rWGS, retaining 388 (85%) with effective treatments. Simulated NBS-rWGS in 454,707 UK Biobank subjects with 29,865 pathogenic or likely pathogenic variants associated with 388 disorders had a true negative rate (specificity) of 99.7% following root cause analysis. In 2,208 critically ill children with suspected genetic disorders and 2,168 of their parents, simulated NBS-rWGS for 388 disorders identified 104 (87%) of 119 diagnoses previously made by rWGS and 15 findings not previously reported (NBS-rWGS negative predictive value 99.6%, true positive rate [sensitivity] 88.8%). Retrospective NBS-rWGS diagnosed 15 children with disorders that had been undetected by conventional NBS. In 43 of the 104 children, had NBS-rWGS-based interventions been started on day of life 5, the Delphi consensus was that symptoms could have been avoided completely in seven critically ill children, mostly in 21, and partially in 13. We invite groups worldwide to refine these NBS-rWGS conditions and join us to prospectively examine clinical utility and cost effectiveness.

An automated 13.5 hour system for scalable diagnosis and acute management guidance for genetic diseases

While many genetic diseases have effective treatments, they frequently progress rapidly to severe morbidity or mortality if those treatments are not implemented immediately. Since front-line physicians frequently lack familiarity with these diseases, timely molecular diagnosis may not improve outcomes. Herein we describe Genome-to-Treatment, an automated, virtual system for genetic disease diagnosis and acute management guidance. Diagnosis is achieved in 13.5 h by expedited whole genome sequencing, with superior analytic performance for structural and copy number variants. An expert panel adjudicated the indications, contraindications, efficacy, and evidence-of-efficacy of 9911 drug, device, dietary, and surgical interventions for 563 severe, childhood, genetic diseases. The 421 (75%) diseases and 1527 (15%) effective interventions retained are integrated with 13 genetic disease information resources and appended to diagnostic reports ( https://gtrx.radygenomiclab.com ). This system provided correct diagnoses in four retrospectively and two prospectively tested infants. The Genome-to-Treatment system facilitates optimal outcomes in children with rapidly progressive genetic diseases.

100,000 Genomes Pilot on Rare-Disease Diagnosis in Health Care - Preliminary Report

BACKGROUND

The U.K. 100,000 Genomes Project is in the process of investigating the role of genome sequencing in patients with undiagnosed rare diseases after usual care and the alignment of this research with health care implementation in the U.K. National Health Service. Other parts of this project focus on patients with cancer and infection.

METHODS

We conducted a pilot study involving 4660 participants from 2183 families, among whom 161 disorders covering a broad spectrum of rare diseases were present. We collected data on clinical features with the use of Human Phenotype Ontology terms, undertook genome sequencing, applied automated variant prioritization on the basis of applied virtual gene panels and phenotypes, and identified novel pathogenic variants through research analysis.

RESULTS

Diagnostic yields varied among family structures and were highest in family trios (both parents and a proband) and families with larger pedigrees. Diagnostic yields were much higher for disorders likely to have a monogenic cause (35%) than for disorders likely to have a complex cause (11%). Diagnostic yields for intellectual disability, hearing disorders, and vision disorders ranged from 40 to 55%. We made genetic diagnoses in 25% of the probands. A total of 14% of the diagnoses were made by means of the combination of research and automated approaches, which was critical for cases in which we found etiologic noncoding, structural, and mitochondrial genome variants and coding variants poorly covered by exome sequencing. Cohortwide burden testing across 57,000 genomes enabled the discovery of three new disease genes and 19 new associations. Of the genetic diagnoses that we made, 25% had immediate ramifications for clinical decision making for the patients or their relatives.

CONCLUSIONS

Our pilot study of genome sequencing in a national health care system showed an increase in diagnostic yield across a range of rare diseases. (Funded by the National Institute for Health Research and others.).

Artificial intelligence enables comprehensive genome interpretation and nomination of candidate diagnoses for rare genetic diseases

BACKGROUND

Clinical interpretation of genetic variants in the context of the patient's phenotype is becoming the largest component of cost and time expenditure for genome-based diagnosis of rare genetic diseases. Artificial intelligence (AI) holds promise to greatly simplify and speed genome interpretation by integrating predictive methods with the growing knowledge of genetic disease. Here we assess the diagnostic performance of Fabric GEM, a new, AI-based, clinical decision support tool for expediting genome interpretation.

METHODS

We benchmarked GEM in a retrospective cohort of 119 probands, mostly NICU infants, diagnosed with rare genetic diseases, who received whole-genome or whole-exome sequencing (WGS, WES). We replicated our analyses in a separate cohort of 60 cases collected from five academic medical centers. For comparison, we also analyzed these cases with current state-of-the-art variant prioritization tools. Included in the comparisons were trio, duo, and singleton cases. Variants underpinning diagnoses spanned diverse modes of inheritance and types, including structural variants (SVs). Patient phenotypes were extracted from clinical notes by two means: manually and using an automated clinical natural language processing (CNLP) tool. Finally, 14 previously unsolved cases were reanalyzed.

RESULTS

GEM ranked over 90% of the causal genes among the top or second candidate and prioritized for review a median of 3 candidate genes per case, using either manually curated or CNLP-derived phenotype descriptions. Ranking of trios and duos was unchanged when analyzed as singletons. In 17 of 20 cases with diagnostic SVs, GEM identified the causal SVs as the top candidate and in 19/20 within the top five, irrespective of whether SV calls were provided or inferred ab initio by GEM using its own internal SV detection algorithm. GEM showed similar performance in absence of parental genotypes. Analysis of 14 previously unsolved cases resulted in a novel finding for one case, candidates ultimately not advanced upon manual review for 3 cases, and no new findings for 10 cases.

CONCLUSIONS

GEM enabled diagnostic interpretation inclusive of all variant types through automated nomination of a very short list of candidate genes and disorders for final review and reporting. In combination with deep phenotyping by CNLP, GEM enables substantial automation of genetic disease diagnosis, potentially decreasing cost and expediting case review.

Diagnosis of genetic diseases in seriously ill children by rapid whole-genome sequencing and automated phenotyping and interpretation

By informing timely targeted treatments, rapid whole-genome sequencing can improve the outcomes of seriously ill children with genetic diseases, particularly infants in neonatal and pediatric intensive care units (ICUs). The need for highly qualified professionals to decipher results, however, precludes widespread implementation. We describe a platform for population-scale, provisional diagnosis of genetic diseases with automated phenotyping and interpretation. Genome sequencing was expedited by bead-based genome library preparation directly from blood samples and sequencing of paired 100-nt reads in 15.5 hours. Clinical natural language processing (CNLP) automatically extracted children's deep phenomes from electronic health records with 80% precision and 93% recall. In 101 children with 105 genetic diseases, a mean of 4.3 CNLP-extracted phenotypic features matched the expected phenotypic features of those diseases, compared with a match of 0.9 phenotypic features used in manual interpretation. We automated provisional diagnosis by combining the ranking of the similarity of a patient's CNLP phenome with respect to the expected phenotypic features of all genetic diseases, together with the ranking of the pathogenicity of all of the patient's genomic variants. Automated, retrospective diagnoses concurred well with expert manual interpretation (97% recall and 99% precision in 95 children with 97 genetic diseases). Prospectively, our platform correctly diagnosed three of seven seriously ill ICU infants (100% precision and recall) with a mean time saving of 22:19 hours. In each case, the diagnosis affected treatment. Genome sequencing with automated phenotyping and interpretation in a median of 20:10 hours may increase adoption in ICUs and, thereby, timely implementation of precise treatments.

The VAAST Variant Prioritizer (VVP): ultrafast, easy to use whole genome variant prioritization tool

Background

Prioritization of sequence variants for diagnosis and discovery of Mendelian diseases is challenging, especially in large collections of whole genome sequences (WGS). Fast, scalable solutions are needed for discovery research, for clinical applications, and for curation of massive public variant repositories such as dbSNP and gnomAD. In response, we have developed VVP, the VAAST Variant Prioritizer. VVP is ultrafast, scales to even the largest variant repositories and genome collections, and its outputs are designed to simplify clinical interpretation of variants of uncertain significance.

Results

We show that scoring the entire contents of dbSNP (> 155 million variants) requires only 95 min using a machine with 4 cpus and 16 GB of RAM, and that a 60X WGS can be processed in less than 5 min. We also demonstrate that VVP can score variants anywhere in the genome, regardless of type, effect, or location. It does so by integrating sequence conservation, the type of sequence change, allele frequencies, variant burden, and zygosity. Finally, we also show that VVP scores are consistently accurate, and easily interpreted, traits not shared by many commonly used tools such as SIFT and CADD.

Conclusions

VVP provides rapid and scalable means to prioritize any sequence variant, anywhere in the genome, and its scores are designed to facilitate variant interpretation using ACMG and NHS guidelines. These traits make it well suited for operation on very large collections of WGS sequences.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2056-y) contains supplementary material, which is available to authorized users.

Collaborating to Compete: Blood Profiling Atlas in Cancer (BloodPAC) Consortium

The cancer community understands the value of blood profiling measurements in assessing and monitoring cancer. We describe an effort among academic, government, biotechnology, diagnostic, and pharmaceutical companies called the Blood Profiling Atlas in Cancer (BloodPAC) Project. BloodPAC will aggregate, make freely available, and harmonize for further analyses, raw datasets, relevant associated clinical data (e.g., clinical diagnosis, treatment history, and outcomes), and sample preparation and handling protocols to accelerate the development of blood profiling assays.

Development of DNA confirmatory and high-risk diagnostic testing for newborns using targeted next-generation DNA sequencing

PURPOSE

Genetic testing is routinely used for second-tier confirmation of newborn sequencing results to rule out false positives and to confirm diagnoses in newborns undergoing inpatient and outpatient care. We developed a targeted next-generation sequencing panel coupled with a variant processing pipeline and demonstrated utility and performance benchmarks across multiple newborn disease presentations in a retrospective clinical study.

METHODS

The test utilizes an in silico gene filter that focuses directly on 126 genes related to newborn screening diseases and is applied to the exome or a next-generation sequencing panel called NBDx. NBDx targets the 126 genes and additional newborn-specific disorders. It integrates DNA isolation from minimally invasive biological specimens, targeted next-generation screening, and rapid characterization of genetic variation.

RESULTS

We report a rapid parallel processing of 8 to 20 cases within 105 hours with high coverage on our NBDx panel. Analytical sensitivity of 99.8% was observed across known mutation hotspots. Concordance calls with or without clinical summaries were 94% and 75%, respectively.

CONCLUSION

Rapid, automated targeted next-generation sequencing and analysis are practical in newborns for second-tier confirmation and neonatal intensive care unit diagnoses, laying a foundation for future primary DNA-based molecular screening of additional disorders and improving existing molecular testing options for newborns.

A unified test of linkage analysis and rare-variant association for analysis of pedigree sequence data

High-throughput sequencing of related individuals has become an important tool for studying human disease. However, owing to technical complexity and lack of available tools, most pedigree-based sequencing studies rely on an ad hoc combination of suboptimal analyses. Here we present pedigree-VAAST (pVAAST), a disease-gene identification tool designed for high-throughput sequence data in pedigrees. pVAAST uses a sequence-based model to perform variant and gene-based linkage analysis. Linkage information is then combined with functional prediction and rare variant case-control association information in a unified statistical framework. pVAAST outperformed linkage and rare-variant association tests in simulations and identified disease-causing genes from whole-genome sequence data in three human pedigrees with dominant, recessive and de novo inheritance patterns. The approach is robust to incomplete penetrance and locus heterogeneity and is applicable to a wide variety of genetic traits. pVAAST maintains high power across studies of monogenic, high-penetrance phenotypes in a single pedigree to highly polygenic, common phenotypes involving hundreds of pedigrees.

Using VAAST to Identify Disease-Associated Variants in Next-Generation Sequencing Data

The VAAST pipeline is specifically designed to identify disease-associated alleles in next-generation sequencing data. In the protocols presented in this paper, we outline the best practices for variant prioritization using VAAST. Examples and test data are provided for case-control, small pedigree, and large pedigree analyses. These protocols will teach users the fundamentals of VAAST, VAAST 2.0, and pVAAST analyses.

An international effort towards developing standards for best practices in analysis, interpretation and reporting of clinical genome sequencing results in the CLARITY Challenge

Background

There is tremendous potential for genome sequencing to improve clinical diagnosis and care once it becomes routinely accessible, but this will require formalizing research methods into clinical best practices in the areas of sequence data generation, analysis, interpretation and reporting. The CLARITY Challenge was designed to spur convergence in methods for diagnosing genetic disease starting from clinical case history and genome sequencing data. DNA samples were obtained from three families with heritable genetic disorders and genomic sequence data were donated by sequencing platform vendors. The challenge was to analyze and interpret these data with the goals of identifying disease-causing variants and reporting the findings in a clinically useful format. Participating contestant groups were solicited broadly, and an independent panel of judges evaluated their performance.

Results

A total of 30 international groups were engaged. The entries reveal a general convergence of practices on most elements of the analysis and interpretation process. However, even given this commonality of approach, only two groups identified the consensus candidate variants in all disease cases, demonstrating a need for consistent fine-tuning of the generally accepted methods. There was greater diversity of the final clinical report content and in the patient consenting process, demonstrating that these areas require additional exploration and standardization.

Conclusions

The CLARITY Challenge provides a comprehensive assessment of current practices for using genome sequencing to diagnose and report genetic diseases. There is remarkable convergence in bioinformatic techniques, but medical interpretation and reporting are areas that require further development by many groups.

Clinical analysis of genome next-generation sequencing data using the Omicia platform

AIMS

Next-generation sequencing is being implemented in the clinical laboratory environment for the purposes of candidate causal variant discovery in patients affected with a variety of genetic disorders. The successful implementation of this technology for diagnosing genetic disorders requires a rapid, user-friendly method to annotate variants and generate short lists of clinically relevant variants of interest. This report describes Omicia's Opal platform, a new software tool designed for variant discovery and interpretation in a clinical laboratory environment. The software allows clinical scientists to process, analyze, interpret and report on personal genome files.

MATERIALS & METHODS

To demonstrate the software, the authors describe the interactive use of the system for the rapid discovery of disease-causing variants using three cases.

RESULTS & CONCLUSION

Here, the authors show the features of the Opal system and their use in uncovering variants of clinical significance.

Integrating precision medicine in the study and clinical treatment of a severely mentally ill person

Background. In recent years, there has been an explosion in the number of technical and medical diagnostic platforms being developed. This has greatly improved our ability to more accurately, and more comprehensively, explore and characterize human biological systems on the individual level. Large quantities of biomedical data are now being generated and archived in many separate research and clinical activities, but there exists a paucity of studies that integrate the areas of clinical neuropsychiatry, personal genomics and brain-machine interfaces. Methods. A single person with severe mental illness was implanted with the Medtronic Reclaim(®) Deep Brain Stimulation (DBS) Therapy device for Obsessive Compulsive Disorder (OCD), targeting his nucleus accumbens/anterior limb of the internal capsule. Programming of the device and psychiatric assessments occurred in an outpatient setting for over two years. His genome was sequenced and variants were detected in the Illumina Whole Genome Sequencing Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory. Results. We report here the detailed phenotypic characterization, clinical-grade whole genome sequencing (WGS), and two-year outcome of a man with severe OCD treated with DBS. Since implantation, this man has reported steady improvement, highlighted by a steady decline in his Yale-Brown Obsessive Compulsive Scale (YBOCS) score from ∼38 to a score of ∼25. A rechargeable Activa RC neurostimulator battery has been of major benefit in terms of facilitating a degree of stability and control over the stimulation. His psychiatric symptoms reliably worsen within hours of the battery becoming depleted, thus providing confirmatory evidence for the efficacy of DBS for OCD in this person. WGS revealed that he is a heterozygote for the p.Val66Met variant in BDNF, encoding a member of the nerve growth factor family, and which has been found to predispose carriers to various psychiatric illnesses. He carries the p.Glu429Ala allele in methylenetetrahydrofolate reductase (MTHFR) and the p.Asp7Asn allele in ChAT, encoding choline O-acetyltransferase, with both alleles having been shown to confer an elevated susceptibility to psychoses. We have found thousands of other variants in his genome, including pharmacogenetic and copy number variants. This information has been archived and offered to this person alongside the clinical sequencing data, so that he and others can re-analyze his genome for years to come. Conclusions. To our knowledge, this is the first study in the clinical neurosciences that integrates detailed neuropsychiatric phenotyping, deep brain stimulation for OCD and clinical-grade WGS with management of genetic results in the medical treatment of one person with severe mental illness. We offer this as an example of precision medicine in neuropsychiatry including brain-implantable devices and genomics-guided preventive health care.

VAAST 2.0: improved variant classification and disease-gene identification using a conservation-controlled amino acid substitution matrix

The need for improved algorithmic support for variant prioritization and disease-gene identification in personal genomes data is widely acknowledged. We previously presented the Variant Annotation, Analysis, and Search Tool (VAAST), which employs an aggregative variant association test that combines both amino acid substitution (AAS) and allele frequencies. Here we describe and benchmark VAAST 2.0, which uses a novel conservation-controlled AAS matrix (CASM), to incorporate information about phylogenetic conservation. We show that the CASM approach improves VAAST’s variant prioritization accuracy compared to its previous implementation, and compared to SIFT, PolyPhen-2, and MutationTaster. We also show that VAAST 2.0 outperforms KBAC, WSS, SKAT, and variable threshold (VT) using published case-control datasets for Crohn disease (NOD2), hypertriglyceridemia (LPL), and breast cancer (CHEK2). VAAST 2.0 also improves search accuracy on simulated datasets across a wide range of allele frequencies, population-attributable disease risks, and allelic heterogeneity, factors that compromise the accuracies of other aggregative variant association tests. We also demonstrate that, although most aggregative variant association tests are designed for common genetic diseases, these tests can be easily adopted as rare Mendelian disease-gene finders with a simple ranking-by-statistical-significance protocol, and the performance compares very favorably to state-of-art filtering approaches. The latter, despite their popularity, have suboptimal performance especially with the increasing case sample size.

A standard variation file format for human genome sequences

Here we describe the Genome Variation Format (GVF) and the 10Gen dataset. GVF, an extension of Generic Feature Format version 3 (GFF3), is a simple tab-delimited format for DNA variant files, which uses Sequence Ontology to describe genome variation data. The 10Gen dataset, ten human genomes in GVF format, is freely available for community analysis from the Sequence Ontology website and from an Amazon elastic block storage (EBS) snapshot for use in Amazon's EC2 cloud computing environment.

Sequence and structural variation in a human genome uncovered by short-read, massively parallel ligation sequencing using two-base encoding

We describe the genome sequencing of an anonymous individual of African origin using a novel ligation-based sequencing assay that enables a unique form of error correction that improves the raw accuracy of the aligned reads to >99.9%, allowing us to accurately call SNPs with as few as two reads per allele. We collected several billion mate-paired reads yielding approximately 18x haploid coverage of aligned sequence and close to 300x clone coverage. Over 98% of the reference genome is covered with at least one uniquely placed read, and 99.65% is spanned by at least one uniquely placed mate-paired clone. We identify over 3.8 million SNPs, 19% of which are novel. Mate-paired data are used to physically resolve haplotype phases of nearly two-thirds of the genotypes obtained and produce phased segments of up to 215 kb. We detect 226,529 intra-read indels, 5590 indels between mate-paired reads, 91 inversions, and four gene fusions. We use a novel approach for detecting indels between mate-paired reads that are smaller than the standard deviation of the insert size of the library and discover deletions in common with those detected with our intra-read approach. Dozens of mutations previously described in OMIM and hundreds of nonsynonymous single-nucleotide and structural variants in genes previously implicated in disease are identified in this individual. There is more genetic variation in the human genome still to be uncovered, and we provide guidance for future surveys in populations and cancer biopsies.

Genome-wide analysis of human disease alleles reveals that their locations are correlated in paralogous proteins

The millions of mutations and polymorphisms that occur in human populations are potential predictors of disease, of our reactions to drugs, of predisposition to microbial infections, and of age-related conditions such as impaired brain and cardiovascular functions. However, predicting the phenotypic consequences and eventual clinical significance of a sequence variant is not an easy task. Computational approaches have found perturbation of conserved amino acids to be a useful criterion for identifying variants likely to have phenotypic consequences. To our knowledge, however, no study to date has explored the potential of variants that occur at homologous positions within paralogous human proteins as a means of identifying polymorphisms with likely phenotypic consequences. In order to investigate the potential of this approach, we have assembled a unique collection of known disease-causing variants from OMIM and the Human Genome Mutation Database (HGMD) and used them to identify and characterize pairs of sequence variants that occur at homologous positions within paralogous human proteins. Our analyses demonstrate that the locations of variants are correlated in paralogous proteins. Moreover, if one member of a variant-pair is disease-causing, its partner is likely to be disease-causing as well. Thus, information about variant-pairs can be used to identify potentially disease-causing variants, extend existing procedures for polymorphism prioritization, and provide a suite of candidates for further diagnostic and therapeutic purposes.

There exists a superabundance of human sequence variations. Testing every sequence variant for association with human disease is often infeasible, as studies must be very large—and hence expensive—to overcome the statistical penalties used to control for multiple tests. A common alternative is to assay only a subset of sequence variants for which there are prior reasons to believe they may be disease-causing. Sequence variants that change conserved amino acids, for example, are often disease-causing. As an adjunct to this approach, we have explored the potential of variants that occur at homologous positions within paralogous human proteins as a means of identifying disease-causing DNA sequence variations. We find that DNA sequence variants co-occur at aligned amino acid pairs more frequently than expected by chance, suggesting that similar functional constraints on paralogous proteins result in coordinated distributions of variants along their lengths. Moreover, if one member of a variant-pair is disease-causing, its partner is likely to be disease-causing as well. These facts provide new avenues for the identification of disease-causing sequence variations.

EGASP: the human ENCODE Genome Annotation Assessment Project

BACKGROUND

We present the results of EGASP, a community experiment to assess the state-of-the-art in genome annotation within the ENCODE regions, which span 1% of the human genome sequence. The experiment had two major goals: the assessment of the accuracy of computational methods to predict protein coding genes; and the overall assessment of the completeness of the current human genome annotations as represented in the ENCODE regions. For the computational prediction assessment, eighteen groups contributed gene predictions. We evaluated these submissions against each other based on a 'reference set' of annotations generated as part of the GENCODE project. These annotations were not available to the prediction groups prior to the submission deadline, so that their predictions were blind and an external advisory committee could perform a fair assessment.

RESULTS

The best methods had at least one gene transcript correctly predicted for close to 70% of the annotated genes. Nevertheless, the multiple transcript accuracy, taking into account alternative splicing, reached only approximately 40% to 50% accuracy. At the coding nucleotide level, the best programs reached an accuracy of 90% in both sensitivity and specificity. Programs relying on mRNA and protein sequences were the most accurate in reproducing the manually curated annotations. Experimental validation shows that only a very small percentage (3.2%) of the selected 221 computationally predicted exons outside of the existing annotation could be verified.

CONCLUSION

This is the first such experiment in human DNA, and we have followed the standards established in a similar experiment, GASP1, in Drosophila melanogaster. We believe the results presented here contribute to the value of ongoing large-scale annotation projects and should guide further experimental methods when being scaled up to the entire human genome sequence.

EGASP: collaboration through competition to find human genes

According to the most recent estimates, the number of human genes is possibly--but not certainly--between 20,000 and 25,000. To contribute strategies to reduce this uncertainty, several groups working on computational gene prediction met recently at the Welcome Trust Sanger Institute with the goal to test and compare predictive methods of genome annotation.

Improved use of SNP information to detect the role of genes

A topical question in genetic association studies is the optimal use of the information provided by genotyped single-nucleotide polymorphisms (SNPs) in order to detect the role of a candidate gene in a multifactorial disease. We propose a strategy called "combination test" that tests the association between a quantitative trait and all possible phased combinations of various numbers of SNPs. We compare this strategy to two alternative strategies: the association test that considers each SNP separately, and a multilocus genotype-based test that considers the phased combination of all SNPs together. To compare these three tests, a quantitative trait was simulated under different models of correspondence between phenotype and genotype, including the extreme case when two SNPs interact with no marginal effects of each SNP. The genotypes were taken from a sample of 290 independent individuals genotyped for three genes with various number of SNPs (from 5-8 SNPs). The results show that the "combination test" is the only one able to detect the association when the two SNPs involved in disease susceptibility interact with no marginal effects. Interestingly, even in the case of a single etiological SNP, the "combination test" performed well. We apply the three tests to Genetic Analysis Workshop 12 (Almasy et al. [2001] Genet. Epidemiol. 21:332-338) simulated data, and show that although there was no interactions between the etiological SNPs, the "combination test" was preferable to the two other compared methods to detect the role of the candidate gene.

Application of a time-delay neural network to promoter annotation in the Drosophila melanogaster genome

Computational methods for automated genome annotation are critical to understanding and interpreting the bewildering mass of genomic sequence data presently being generated and released. A neural network model of the structural and compositional properties of a eukaryotic core promoter region has been developed and its application for analysis of the Drosophila melanogaster genome is presented. The model uses a time-delay architecture, a special case of a feed-forward neural network. The structure of this model allows for variable spacing between functional binding sites, which is known to play a key role in the transcription initiation process. Application of this model to a test set of core promoters not only gave better discrimination of potential promoter sites than previous statistical or neural network models, but also revealed indirectly subtle properties of the transcription initiation signal. When tested in the Adh region of 2.9 Mbases of the Drosophila genome, the neural network for promoter prediction (NNPP) program that incorporates the time-delay neural network model gives a recognition rate of 75% (69/92) with a false positive rate of 1/547 bases. The present work can be regarded as one of the first intensive studies that applies novel gene regulation technologies to the identification of the complex gene regulation sites in the genome of Drosophila melanogaster.

Annotating eukaryote genomes

The Genome Annotation Assessment Project tested current methods of gene identification, including a critical assessment of the accuracy of different methods. Two new databases have provided new resources for gene annotation: these are the InterPro database of protein domains and motifs, and the Gene Ontology database for terms that describe the molecular functions and biological roles of gene products. Efforts in genome annotation are most often based upon advances in computer systems that are specifically designed to deal with the tremendous amounts of data being generated by current sequencing projects. These efforts in analysis are being linked to new ways of visualizing computationally annotated genomes.

Genie--gene finding in Drosophila melanogaster

A hidden Markov model-based gene-finding system called Genie was applied to the genomic Adh region in Drosophila melanogaster as a part of the Genome Annotation Assessment Project (GASP). Predictions from three versions of the Genie gene-finding system were submitted, one based on statistical properties of coding genes, a second included EST alignment information, and a third that integrated protein sequence homology information. All three programs were trained on the provided Drosophila training data. In addition, promoter assignments from an integrated neural network were submitted. The gene assignments overlapped >90% of the 222 annotated genes and 26 possibly novel genes were predicted, of which some might be overpredictions. The system correctly identified the exon boundaries of 70% of the exons in cDNA-confirmed genes and 77% of the exons with the addition of EST sequence alignments. The best of the three Genie submissions predicted 19 of the annotated 43 gene structures entirely correct (44%). In the promoter category, only 30% of the transcription start sites could be detected, but by integrating this program as a sensor into Genie the false-positive rate could be dropped to 1/16,786 (0.006%). The results of the experiment on the long contiguous genomic sequence revealed some problems concerning gene assembly in Genie. The results were used to improve the system. We show that Genie is a robust hidden Markov model system that allows for a generalized integration of information from different sources such as signal sensors (splice sites, start codon, etc.), content sensors (exons, introns, intergenic) and alignments of mRNA, EST, and peptide sequences. The assessment showed that Genie could effectively be used for the annotation of complete genomes from higher organisms.

Genome annotation assessment in Drosophila melanogaster

Computational methods for automated genome annotation are critical to our community's ability to make full use of the large volume of genomic sequence being generated and released. To explore the accuracy of these automated feature prediction tools in the genomes of higher organisms, we evaluated their performance on a large, well-characterized sequence contig from the Adh region of Drosophila melanogaster. This experiment, known as the Genome Annotation Assessment Project (GASP), was launched in May 1999. Twelve groups, applying state-of-the-art tools, contributed predictions for features including gene structure, protein homologies, promoter sites, and repeat elements. We evaluated these predictions using two standards, one based on previously unreleased high-quality full-length cDNA sequences and a second based on the set of annotations generated as part of an in-depth study of the region by a group of Drosophila experts. Although these standard sets only approximate the unknown distribution of features in this region, we believe that when taken in context the results of an evaluation based on them are meaningful. The results were presented as a tutorial at the conference on Intelligent Systems in Molecular Biology (ISMB-99) in August 1999. Over 95% of the coding nucleotides in the region were correctly identified by the majority of the gene finders, and the correct intron/exon structures were predicted for >40% of the genes. Homology-based annotation techniques recognized and associated functions with almost half of the genes in the region; the remainder were only identified by the ab initio techniques. This experiment also presents the first assessment of promoter prediction techniques for a significant number of genes in a large contiguous region. We discovered that the promoter predictors' high false-positive rates make their predictions difficult to use. Integrating gene finding and cDNA/EST alignments with promoter predictions decreases the number of false-positive classifications but discovers less than one-third of the promoters in the region. We believe that by establishing standards for evaluating genomic annotations and by assessing the performance of existing automated genome annotation tools, this experiment establishes a baseline that contributes to the value of ongoing large-scale annotation projects and should guide further research in genome informatics.

The genome sequence of Drosophila melanogaster

The fly Drosophila melanogaster is one of the most intensively studied organisms in biology and serves as a model system for the investigation of many developmental and cellular processes common to higher eukaryotes, including humans. We have determined the nucleotide sequence of nearly all of the approximately 120-megabase euchromatic portion of the Drosophila genome using a whole-genome shotgun sequencing strategy supported by extensive clone-based sequence and a high-quality bacterial artificial chromosome physical map. Efforts are under way to close the remaining gaps; however, the sequence is of sufficient accuracy and contiguity to be declared substantially complete and to support an initial analysis of genome structure and preliminary gene annotation and interpretation. The genome encodes approximately 13,600 genes, somewhat fewer than the smaller Caenorhabditis elegans genome, but with comparable functional diversity.

An exploration of the sequence of a 2.9-Mb region of the genome of Drosophila melanogaster: the Adh region

A contiguous sequence of nearly 3 Mb from the genome of Drosophila melanogaster has been sequenced from a series of overlapping P1 and BAC clones. This region covers 69 chromosome polytene bands on chromosome arm 2L, including the genetically well-characterized "Adh region." A computational analysis of the sequence predicts 218 protein-coding genes, 11 tRNAs, and 17 transposable element sequences. At least 38 of the protein-coding genes are arranged in clusters of from 2 to 6 closely related genes, suggesting extensive tandem duplication. The gene density is one protein-coding gene every 13 kb; the transposable element density is one element every 171 kb. Of 73 genes in this region identified by genetic analysis, 49 have been located on the sequence; P-element insertions have been mapped to 43 genes. Ninety-five (44%) of the known and predicted genes match a Drosophila EST, and 144 (66%) have clear similarities to proteins in other organisms. Genes known to have mutant phenotypes are more likely to be represented in cDNA libraries, and far more likely to have products similar to proteins of other organisms, than are genes with no known mutant phenotype. Over 650 chromosome aberration breakpoints map to this chromosome region, and their nonrandom distribution on the genetic map reflects variation in gene spacing on the DNA. This is the first large-scale analysis of the genome of D. melanogaster at the sequence level. In addition to the direct results obtained, this analysis has allowed us to develop and test methods that will be needed to interpret the complete sequence of the genome of this species. Before beginning a Hunt, it is wise to ask someone what you are looking for before you begin looking for it. Milne 1926

Interpolated markov chains for eukaryotic promoter recognition

MOTIVATION

We describe a new content-based approach for the detection of promoter regions of eukaryotic protein encoding genes. Our system is based on three interpolated Markov chains (IMCs) of different order which are trained on coding, non-coding and promoter sequences. It was recently shown that the interpolation of Markov chains leads to stable parameters and improves on the results in microbial gene finding (Salzberg et al., Nucleic Acids Res., 26, 544-548, 1998). Here, we present new methods for an automated estimation of optimal interpolation parameters and show how the IMCs can be applied to detect promoters in contiguous DNA sequences. Our interpolation approach can also be employed to obtain a reliable scoring function for human coding DNA regions, and the trained models can easily be incorporated in the general framework for gene recognition systems.

RESULTS

A 5-fold cross-validation evaluation of our IMC approach on a representative sequence set yielded a mean correlation coefficient of 0.84 (promoter versus coding sequences) and 0.53 (promoter versus non-coding sequences). Applied to the task of eukaryotic promoter region identification in genomic DNA sequences, our classifier identifies 50% of the promoter regions in the sequences used in the most recent review and comparison by Fickett and Hatzigeorgiou ( Genome Res., 7, 861-878, 1997), while having a false-positive rate of 1/849 bp.

Integrating database homology in a probabilistic gene structure model

We present an improved stochastic model of genes in DNA, and describe a method for integrating database homology into the probabilistic framework. A generalized hidden Markov model (GHMM) describes the grammar of a legal parse of a DNA sequence. Probabilities are estimated for gene features by using dynamic programming to combine information from multiple sensors. We show how matches to homologous sequences from a database can be integrated into the probability estimation by interpreting the likelihood of a sequence in terms of the bit-cost to encode a sequence given a homology match. We also demonstrate how homology matches in protein databases can be exploited to help identify splice sites. Our experiments show significant improvements in the sensitivity and specificity of gene structure identification when these new features are added to our gene-finding system, Genie. Experimental results in tests using a standard set of annotated genes showed that Genie identified 95% of coding nucleotides correctly with a specificity of 91%, and 77% of exons were identified exactly.

Improved splice site detection in Genie

We present an improved splice site predictor for the genefinding program Genie. Genie is based on a generalized Hidden Markov Model (GHMM) that describes the grammar of a legal parse of a multi-exon gene in a DNA sequence. In Genie, probabilities are estimated for gene features by using dynamic programming to combine information from multiple content and signal sensors, including sensors that integrate matches to homologous sequences from a database. One of the hardest problems in genefinding is to determine the complete gene structure correctly. The splice site sensors are the key signal sensors that address this problem. We replaced the existing splice site sensors in Genie with two novel neural networks based on dinucleotide frequencies. Using these novel sensors, Genie shows significant improvements in the sensitivity and specificity of gene structure identification. Experimental results in tests using a standard set of annotated genes showed that Genie identified 86% of coding nucleotides correctly with a specificity of 85%, versus 80% and 84% in the older system. In further splice site experiments, we also looked at correlations between splice site scores and intron and exon lengths, as well as at the effect of distance to the nearest splice site on false positive rates.

Distance distributions in proteins: a six-parameter representation

We present a statistical analysis of protein structures based on interatomic C alpha distances. The overall distance distributions reflect in detail the contents of sequence-specific substructures maintained by local interactions (such as alpha-helixes) and longer range interactions (such as disulfide bridges and beta-sheets). We also show that a volume scaling of the distances makes distance distributions for protein chains of different length superimposable. Distance distributions were also calculated specifically for amino acids separated by a given number of residues. Specific features in these distributions are visible for sequence separations of up to 20 amino acid residues. A simple representation, which preserves most of the information in the distance distributions, was obtained using six parameters only. The parameters give rise to canonical distance intervals and when predicting coarse-grained distance constraints by methods such as data-driven artificial neural networks, these should preferably be selected from these intervals. We discuss the use of the six parameters for determining or reconstructing 3-D protein structures.

A generalized hidden Markov model for the recognition of human genes in DNA

We present a statistical model of genes in DNA. A Generalized Hidden Markov Model (GHMM) provides the framework for describing the grammar of a legal parse of a DNA sequence (Stormo & Haussler 1994). Probabilities are assigned to transitions between states in the GHMM and to the generation of each nucleotide base given a particular state. Machine learning techniques are applied to optimize these probabilities using a standardized training set. Given a new candidate sequence, the best parse is deduced from the model using a dynamic programming algorithm to identify the path through the model with maximum probability. The GHMM is flexible and modular, so new sensors and additional states can be inserted easily. In addition, it provides simple solutions for integrating cardinality constraints, reading frame constraints, "indels", and homology searching. The description and results of an implementation of such a gene-finding model, called Genie, is presented. The exon sensor is a codon frequency model conditioned on windowed nucleotide frequency and the preceding codon. Two neural networks are used, as in (Brunak, Engelbrecht, & Knudsen 1991), for splice site prediction. We show that this simple model performs quite well. For a cross-validated standard test set of 304 genes [ftp:@www-hgc.lbl.gov/pub/genesets] in human DNA, our gene-finding system identified up to 85% of protein-coding bases correctly with a specificity of 80%. 58% of exons were exactly identified with a specificity of 51%. Genie is shown to perform favorably compared with several other gene-finding systems.

Automated measurement of serum thyroxine with the "ARIA II," as compared with competitive protein binding and radioimmunoassay

Two conventional serum thyroxine assays, run in separate laboratories, one by competitive protein binding and one by radioimmunoassay, were used to evaluate the automated ARIA II (Becton Dickinson Immunodiagnostics) serum thyroxine assay. Competitive protein binding as compared to ARIA II with 111 clinical serum samples gave a slope of 1.04 and a correlation coefficient of 0.94. The radioimmunoassay comparison to ARIA II with 53 clinical serum samples gave a slope of 1.05 and a correlation coefficient of 0.92. The ARIA II inter-assay coefficient of variation for 10 replicates of low, medium, and high thyroxine serum samples was 6.2, 6.0, and 2.9%, respectively, with an inter-assay coefficient of variation among 15 different assays of 15.5, 10.1, and 7.9%. The automated ARIA II, with a 2.2-min cycle per sample, gives results that compare well with those by manual methodology.

Automated measurement of serum thyroxine with the "ARIA II," as compared with competitive protein binding and radioimmunoassay

Two conventional serum thyroxine assays, run in separate laboratories, one by competitive protein binding and one by radioimmunoassay, were used to evaluate the automated ARIA II (Becton Dickinson Immunodiagnostics) serum thyroxine assay. Competitive protein binding as compared to ARIA II with 111 clinical serum samples gave a slope of 1.04 and a correlation coefficient of 0.94. The radioimmunoassay comparison to ARIA II with 53 clinical serum samples gave a slope of 1.05 and a correlation coefficient of 0.92. The ARIA II inter-assay coefficient of variation for 10 replicates of low, medium, and high thyroxine serum samples was 6.2, 6.0, and 2.9%, respectively, with an inter-assay coefficient of variation among 15 different assays of 15.5, 10.1, and 7.9%. The automated ARIA II, with a 2.2-min cycle per sample, gives results that compare well with those by manual methodology.