Clinical Analysis and Interpretation of Cancer Genome Data

Abstract

The scale of tumor genomic profiling is rapidly outpacing human cognitive capacity to make clinical decisions without the aid of tools. New frameworks are needed to help researchers and clinicians process the information emerging from the explosive growth in both the number of tumor genetic variants routinely tested and the respective knowledge to interpret their clinical significance. We review the current state, limitations, and future trends in methods to support the clinical analysis and interpretation of cancer genomes. This includes the processes of genome-scale variant identification, including tools for sequence alignment, tumor–germline comparison, and molecular annotation of variants. The process of clinical interpretation of tumor variants includes classification of the effect of the variant, reporting the results to clinicians, and enabling the clinician to make a clinical decision based on the genomic information integrated with other clinical features. We describe existing knowledge bases, databases, algorithms, and tools for identification and visualization of tumor variants and their actionable subsets. With the decreasing cost of tumor gene mutation testing and the increasing number of actionable therapeutics, we expect the methods for analysis and interpretation of cancer genomes to continue to evolve to meet the needs of patient-centered clinical decision making. The science of computational cancer medicine is still in its infancy; however, there is a clear need to continue the development of knowledge bases, best practices, tools, and validation experiments for successful clinical implementation in oncology.

INTRODUCTION

As the amount of information generated by an individual patient's tumor genomic profiling increases, clinicians will no longer be able to analyze, interpret, and act on this information without tools that effectively mine the data for clinical use. This remarkable increase in scale is due to the increasing number of altered genes that may portend clinical impact, the larger number of gene variants being routinely tested in some tumor types,^1–3 and more generally the explosive growth in clinical knowledge linking somatic or germline variants to predictive or prognostic information. In the future, providers across the health care system—from molecular pathologists and oncologists to genetic counselors—will be challenged to rapidly analyze and interpret the results of emerging genomic technologies for effective patient care.

The clinical workflow for analysis and interpretation of the cancer genome (Fig 1) begins with the genome testing modality, that is, the technology or platform used to generate raw genetic or molecular data (modality options are reviewed in detail elsewhere⁴). After the assay has been performed, the molecular pathologist identifies alterations in the tumor. Depending on the assay modality, variant identification may be done by visual inspection by a trained specialist (eg, ALK fusion detection assays⁵) or require the assistance of automated algorithms (eg, assays that assess the entire sequence of one or more genes⁶). Automated methods for variant identification consist of several processing steps, including algorithms for sequence alignment that take as input a reference sequence, algorithms for comparison of the tumor and normal genome, and molecular annotation of variants. As the scale of data rises from a few well-defined, commonly mutated alterations (hotspots) to individual whole genes to entire genomes, several factors make analysis increasingly difficult. These include artifacts related to sequencing, overall breadth of coverage across one or more genes, differences in depth of coverage at a particular site that affect the power to detect variants, and variability in the reference genome data set as it relates to an individual patient.

Fig 1.

Clinical workflow for tumor genome analysis and interpretation. The workflow begins with the genome-testing modality that may or may not have digital output. Variant identification may be done manually or with the assistance of automated algorithms, depending on the modality. Automated methods for variant identification include several processes shown at the top, including algorithms for sequence alignment that take as input a reference sequence, algorithms for comparison of the tumor and normal genome, and variant annotation. Clinical interpretation of the variant is a process that includes a determination of the size of variant effect and its interaction with other variants as well as an analysis of the strength of the evidence of the effect. These processes require the use of knowledge bases of variant–drug–disease relationships. Actionable results are reported as well as variants of unknown significance (VUS). The process culminates with a clinician using the information to make a clinical decision.

The clinical interpretation of variants identified by genomic profiling is a process that includes classifying the clinical effect of the variant and reporting the results in ways that are meaningful to practicing oncologists. The need for robust clinical interpretation requires considerations that are distinct from those of both the assay testing modality and the number of genes or variants assessed. Unlike many other clinical assays, multiplex or sequencing-based tumor testing may result in many simultaneous and significant outcomes per gene that occur in any given tumor sample. Furthermore, various genetic alterations may predict either sensitivity or resistance (intrinsic or acquired) to a particular drug. For example, in non–small-cell lung cancer, some EGFR mutations are associated with sensitivity to small-molecule epidermal growth factor receptor (EGFR) antagonists (eg, EGFR L858R mutation⁷), whereas others confer primary or secondary resistance to the same drugs (eg, EGFR T790M⁸). Furthermore, there are even more variants in EGFR (and most other cancer genes) whose clinical significance is currently unknown; that is, there exist no definitive clinical data to suggest sensitivity or resistance to EGFR-targeted therapy for tumors harboring those particular EGFR alterations.^9,10 Similarly, mutations that predict responsiveness to a therapy in some contexts (eg, RAF inhibitors in BRAF[V600]-mutant melanoma) may be associated wtih entirely different clinical interpretations in others (eg, RAF inhibitors in BRAF[V600]-mutant colorectal carcinoma¹¹). In the end, the physician must interpret the utility of tumor genomic results in the context of additional clinical, histologic, and molecular information to make an informed clinical decision.

With hundreds to thousands of tumor variants observed in the coding region of an individual's genome and thousands to millions of variants across an individual's cancer genome,¹² it will not be possible for molecular pathologists and oncologists to identify and appropriately annotate the clinical significance of each variant by manually assigning an individual interpretation to the variants found in every patient. Although the computational algorithms and knowledge resources available today to assist this process are still in their infancy, they demonstrate great potential for realizing this aim and thus paving the way for widespread applications of genomic profiling in clinical oncology.

THE PROCESS OF GENOME-SCALE VARIANT IDENTIFICATION

Current clinical genomic testing typically focuses on one or two well-defined, targetable alterations (hotspots) in a small number of genes, and the binary presence or absence of single mutational events is often sufficient to guide clinical decision making. Existing research-grade algorithms for revealing biologically significant alterations within the high volume of genomics data, although powerful for discovery purposes, are not sufficient for clinical deployment. To bring genome-scale data to the clinic, the computational process for variant identification (variant calling) requires adaptations of existing tools and the development of new annotation methods to allow for effective downstream interpretation of these variants.

Sequence Alignment

Large-scale genomics platforms (eg, those that use massively parallel sequencing) produce data reflecting the raw sequence reads that emerge from the sequencing instrument used by the clinical platform. Regardless of whether that output consists of the coding regions of a subset of genes (targeted sequencing), the entire coding regions of the genome (exome), the entire genome, or the spectrum of RNA transcription across the genome (transcriptome), the first step in computational analysis is aligning the raw data to the reference human genome to better understand variations (Fig 2). Multiple tools exist for this process^13–21 and can be adapted for clinical use. Critically, depending on the genomic territory probed with the sequencing modality, sequence alignment can still be a computationally intensive and therefore time-consuming process that may challenge existing clinical informatics infrastructures to ensure clinically useful data turnaround. In addition, although cross-comparative analysis of multiple aligners for research purposes has been initiated,²² determination of the most effective clinical alignment algorithm has not yet been established. Because alignment to the reference genome is the foundation for all downstream analysis, clinically focused cross-comparator studies are needed to determine the optimal alignment approach, as these may inform understanding of the sensitivity and specificity of such approaches.

Fig 2.

A representative set of tools for the analysis and interpretation of genome sequencing data. These include (A) a listing of representative algorithms for sequencing alignment, (B) variant identification, (C) variant annotation, and (D) clinical interpretation. Boldfaced entries are those specifically geared toward tumor versus normal analysis.

Variant Identification and Tumor/Germline Comparison

After sequence alignment, several analytic tools are available to identify variants in the tumor genome. Targeted gene panels that focus on a subset of highly characterized tumor genes may proceed directly to variant identification without analysis relative to the patient's germline data because well-defined germline events may be filtered out.^6,23 However, such avenues may have difficulty in distinguishing some private or low-frequency germline alterations from bona fide somatic alterations in tumors; parallel analysis of paired germline DNA is needed to enable the maximal power to distinguish true somatic events. The variant identification algorithms developed for exome or genome-wide analyses to nominate putative mutations,^24–31 small insertions/deletions (“indels”),^30,32–37 and copy number alterations^29,38–44) can be adapted for targeted panels (Fig 2).

As noted previously, effective variant identification for an entire tumor exome or genome always requires direct comparison of the tumor genome to the patient's germline to identify somatic variants. Many variant identification software packages for exome or genome-level data are specifically primed for use in clinical cancer care (Fig 2). However, clinical-grade sequencing may require additional quality-assurance methods to ensure that these variant identification algorithms do not accidentally filter out known, clinically significant variants. Furthermore, identifying the high-confidence absence of expected or likely tumor variants that have clinical implications (pertinent negatives) will be equally relevant for subsequent clinical interpretation and represents an essential component of clinical variant identification tasks.

Reliable detection of clinically significant rearrangements (eg, ALK fusions) can be difficult without genome or transcriptome-level data, because targeted panels or exome sequencing are unlikely to capture all intronic regions where rearrangements may occur. In the future, potential solutions may include concomitant transcriptome sequencing to capture novel fusion transcripts^45,46 or intentional tiling of baits across high-yield intronic regions as part of a targeted gene panel.²³ Analysis tools to reliably identify fusions in such regions exist and may be deployed for clinical use once the appropriate platform is chosen.^32,47,48

Molecular Annotation of Variants

The somatic events observed with analysis tools are typically represented in computer files as genomic coordinates with allele changes or segments of copy number gain or loss. To proceed with effective clinical interpretation of these events, translation of these data for human use with effective molecular annotation is necessary. Publically available annotation tools exist to convert these data into formats that use gene names and protein changes based on established public resources.^49–54 Salient sequencing data metrics may include alternative transcripts expressed from query loci, locus-specific coverage and the variant allelic fraction (defined as the number of alternate reads at the site divided by the total number of reads at that site). Additional resources may link genetic alterations to other databases that can aid downstream clinical interpretation, including the predicted effect of the variant on the protein^52,55,56 or the frequency of this event in published cancer genomics research studies.⁵⁷ At the present time, these annotations are typically focused on research-oriented pursuits, and new databases will be needed to frame clinically oriented molecular annotation.

THE PROCESS OF CLINICAL INTERPRETATION OF TUMOR VARIANTS

Once all tumor variants in a patient's genome have been identified, clinical interpretation of each variant is needed to identify the subset that may affect medical decision making. The process of clinical interpretation includes classification of the effect of the variant, reporting the results to clinicians, and enabling the physician to make a management decision based on the genomic information integrated with other clinical features.

Classifying the Clinical Effects of Genomic Variants

Interpretation of the clinical effect of a tumor variant includes classifying (1) the type of clinical effect, (2) the strength of evidence of the effect, and (3) the size of the effect. Framing observed variants in this manner facilitates their clinical interpretation by focusing attention on variants of different types with the most size and strength of effect for an individual patient (Fig 3).

Fig 3.

Classification of the clinical effect of the variant taking into account the type of effect, strength of the evidence, and the size of the effect. Variants in the top right quadrant should have the highest priority with respect to actionable clinical decisions.

Classification of the type of clinical effect.

As with other clinical biomarkers, cancer-related gene variants can be classified by their type of clinical effect as biomarkers that predict risk of disease, confirm a diagnosis, predict prognosis, predict response or resistance to treatment, or measure response to treatment.⁵⁸ For example, deleterious germline BRCA1 mutations confer increased risk of breast cancer,⁵⁹ IDH1 mutations may predict favorable prognosis in gliomas,⁶⁰ BRAF mutations may predict sensitivity to BRAF inhibitors in melanoma,⁶¹ KRAS mutations may predict resistance to anti-EGFR antibody treatment,⁶² and NPM1 is used to measure minimal residual disease in acute myeloid leukemia.⁶³ Some tumor variants serve multiple clinical utilities. For example, BCR-ABL translocations are diagnostic, prognostic, and predictive of response to treatment with imatinib,⁶⁴ defining a distinct clinical subtype of chronic myelogenous leukemia.

Likewise, the clinical effect may be influenced by multiple variants that may be operant simultaneously. In acute myeloid leukemia, for example, the combination of certain NPM1 mutations without concurrent FLT3 mutations has been associated with complete remission and favorable outcomes.⁶⁵ Similarly, multiple EGFR mutations may occur simultaneously in EGFR mutant lung adenocarcinoma that confer opposite sensitivities to EGFR tyrosine kinase inhibitors.⁶⁶ Initial efforts to computationally represent these complex relationships are underway, including one approach to systematically annotate genomic relationships linked to clinical actions and build a series of heuristics to score the events relative to one another.⁶⁷

Classification of the strength of evidence.

Large-scale profiles of tumor gene alterations present several challenges with respect to classifying and communicating the strength of the evidence underlying a given clinical effect. Strength of evidence is based on the maturity, quality, and type of studies that inform the potential clinical utility of a genomic alteration.⁶⁸ When reporting clinical significance, it is not sufficient to simply describe the clinical effect with respect to disease and treatment. More detailed information on the evidence for those relationships is typically required for clinical decision making.

Evidence-based approaches to biomarker testing for routine clinical care limit testing to biomarkers with well-established clinical utility. Multiple frameworks have been proposed that classify the evidence related to the clinical utility of tumor biomarkers, including: (1) the Tumor Marker Utility Grading Systems,⁶⁸ (2) levels of evidence for using archived tissue to determine the clinical validity of tumor markers,⁶⁹ (3) levels of evidence combined with tiers to represent different clinical scenarios for a genomic event (eg, sensitivity, resistance, or prognostic/diagnostic effect),⁶ and (4) the National Comprehensive Cancer Network categories of evidence and consensus.⁷⁰ These frameworks provide a consistent language for clinical practice guidelines to make recommendations on the use of a biomarker or biomarker-driven therapy in clinical practice. At the present time, however, relatively few tumor genes meet criteria for routine testing.⁷⁰

As the cost of deep sequencing drops and the clinical utility of multiple tumor biomarkers is established, it should become increasingly cost effective to perform large-scale cancer gene mutation profiling routinely. The resulting proliferation of cancer genomic data may present specific challenges with respect to communication of the strength of evidence for the clinical effects to physicians and patients. The preceding frameworks rely on clinical evidence from relatively large populations of patients. Such recommendations are feasible with variants of relatively high prevalence; however, many tumor variants occur with a frequency of less than 1% in a particular subpopulation. For instance, more than 100 different primary and secondary EGFR mutations have been described in literature case reports for patients with non–small-cell lung cancer.⁵⁷ Many of these variants occur with extremely low frequency and are of unknown clinical significance. Routine testing of the full gene sequence as opposed to testing strategies that only assess for high-prevalence variants are likely to identify more of these rare variants of unknown significance in tumor genes that have well-established clinical utility.

Thus, as genomic testing expands to large numbers of genes, there will often be no prospective clinical trials with sufficient power to determine the clinical significance of rare variants. Yet clinicians will still need assistance to interpret the clinical significance of these variants. One group describes the evolving evidence perspective leading toward patient-centered outcomes research.⁷¹ This framework describes levels of evidence ranging from expert opinion (N of 1), evidence-based medicine (N of many), comparative effectiveness research (N of many), and patient-centered outcomes research (many N′s of 1). The idea is that for variants of unknown significance, providers will make recommendations to individual patients based on as many similar patients that can be found. When such clinical data are not available, predictive modeling or preclinical data may be the only evidence of possible clinical effect. Such information may be useful to guide clinical trial selection in certain research settings, but as yet there exists no validated framework to incorporate preclinical evidence into clinical decision making regarding the use of tumor genomic data. Going forward, a deeper biologic understanding of targetable pathways in the appropriate cell or lineage contexts will likely also play a key role in identifying clinically relevant variants.

Classification of the size of the effect.

The size of the effect relates to the degree of benefit or harm a biomarker predicts.⁶⁸ The clinical vernacular for classifying tumor variants based on the type and size of clinical effect is rapidly evolving. For example, “EGFR mutations sensitive to EGFR tyrosine kinase inhibitors” has become a common description in clinical trial eligibility criteria. It refers to the known subgroup of EGFR mutations that are sensitive to drugs such as erlotinib and gefitinib. Binary classifications such as “sensitive” and “resistant” are becoming more common in describing biomarker-related drug effects. Other frameworks describe the size of the clinical effect on a continuum from likely benefit versus possible harm.⁵⁷ Such frameworks use phrases on a continuum of size of effect such as “treatment should be administered,” “it is reasonable to consider treatment,” “treatment may be considered,” and “treatment should not be administered.” Others have adopted “+” versus “–” classification of the size of effects such as the framework used in the well-established Sanford Guide for Antimicrobials,⁷² which uses “+++” to describe the relative sensitivity of an antimicrobial to treat a particular microorganism. The clinical vernacular for communicating the effect size of a tumor variant specifically with respect to treatment choice is still evolving, but it is expected that binary classifications will not capture the nuances of effect size as more data about variants within a patient's genome become available in preclinical and clinical studies.

Reporting Results

Whether testing a single gene variant or the entire cancer genome, multiple challenges remain in discerning the optimal means to report results in ways that are clinically useful for practicing oncologists. Open-ended issues for the clinical audience include (1) what to report, (2) how to format the report, and (3) if, when, and how to notify providers when new clinical evidence emerges.

Choosing what to report.

After observed variants have been comprehensively classified with clinical applicability in mind, they must be reported to the clinician in a manner that enables facile decision support. Multiple approaches to reporting exist, reflecting a spectrum of data access or restriction. Approaches favoring full access report all variants and allow clinicians complete control over what actions to take. Approaches favoring restricted access only report variants with well-established size and strength of effects. Overall, the challenge with reporting is to balance the need for deep exploration of a tumor genome while simultaneously being user-friendly to allow for efficient assessment of clinically pertinent results. It should be noted that the optimal return of genetic results to clinicians and patients also remains an area of active investigation from an ethical, legal, and social perspective.^73–75

Formatting the report for a clinical audience.

Because tumor variants may be classified in multiple ways, it is important to organize the content of the reports in multiple ways to facilitate clinical decisions of different types. One common framework is a gene-oriented approach in which genomic results are organized by their type of clinical effect (see Classification of the Type of Clinical Effect, earlier). When large numbers of genes are assessed simultaneously, clinical decision making may be aided by segregating the results based on the type of clinical action that may follow. For instance, reports of assays that assess both tumor and germline variants should draw attention to variants that may reveal a hereditary cancer syndrome in a separate section from those that confer sensitivity or resistance to a particular therapy. Germline variants unrelated to cancer that have clear medical implications for either the patient or her family should also be noted.

Likewise, it is useful to organize treatment-related results by the drugs they may affect. This is especially important in the context of genomic aberrations that confer resistance to drugs, because they can co-occur with aberrations that confer sensitivity to the same drug (eg, EGFR T790M).⁸ A drug-oriented view of the results can thus take into account multiple findings simultaneously. Such views are often grouped based on those drugs that are likely to be effective and those that are not likely to be effective. This simplifying approach can be very helpful for clinical decision making; however, care must be taken to ensure that off-label drug associations are clearly stated.

From this perspective, reports of tumor genomic profiles should include an assessment of the level of evidence and strength of the effect with reference to the primary literature. Ideally, content should likewise be formatted in such a way as to draw the clinician's attention to those results with the highest level of evidence and strength of effect (Fig 3).

Several groups that perform tumor mutation analysis are convening molecular tumor boards before finalizing the contents of the interpretive report (eg, Roychowdhury et al³ and Tran et al ⁷⁶). Other groups that perform tumor genomic analysis on a fixed number of cancer genes have developed electronic versions of their reports that link to outside knowledge resources.^2,77 This interactive reporting strategy has the advantage of reporting the analytic validity of an assay at a moment in time, while allowing the clinical utility of the results to continuously be updated by outside and centrally maintained sources. Several tools in use⁷⁸ or under development⁶⁷ enable molecular pathologists to more easily create interpretive reports with the frameworks described previously. They are being designed to leverage the information from multiple publically available web-based resources, such that as much information as desired can be obtained with ease and within the framework of an electronic medical record.

The knowledge resources available today vary in their degree of clinical applicability, as many were originally created to facilitate cancer research. They also vary in their scope and degree of coverage of the cancer domain. Some only cover information at the gene level, whereas others cover information at the gene variant level. Some focus on gene–disease relationships and others on gene–drug relationships, which correspond to the types of clinic effects (eg, diagnosis v predict response to treatment) that can be inferred from the knowledge. Finally, these resources vary in the levels of evidence that they cover and their public accessibility. Table 1 describes the cancer-specific databases and knowledge bases for tumor variant information that are potentially clinically applicable. Table 2 describes general gene–drug databases that also may provide some utility for the interpretation of the clinical significance of tumor variants.

Table 1.

Cancer-Specific Knowledge Bases to Assist in Clinical Interpretation of Tumor Variants

Knowledge Base or Database	Description	Types of Relationships	Clinical Domains	Size of Database (2012)
Catalogue of Somatic Mutations in Cancer (COSMIC)57	COSMIC is a database of somatic mutations in cancers derived from curation of case reports in the literature	Tumor variant–disease	All cancer diagnoses	> 700,000 tumors cases with > 280,000 mutations
Genomics of Drug Sensitivity in Cancer79	The Genomics of Drug Sensitivity in Cancer database contains information on the drug sensitivity of cancer cell lines	Tumor gene–drug	Cancer cell lines	Screen of several hundred cancer cell lines with 130 cancer drugs
The Cancer Genome Atlas (TCGA) data portal80	The TCGA data portal contains cases of tumor with extensive molecular annotations and more limited clinical annotations	Tumor variant–disease–drug–outcome	21 cancer types	21 cancer diagnoses with > 6,000 tumor samples with genome and clinical annotations
DNA-Mutation Inventory to Refine and Enhance Cancer Treatment (DIRECT)81	DIRECT is a database of cases of non–small-cell lung cancer reported in the literature with tumor variant(s), drug exposure, and clinical outcome annotations25	Tumor variant-disease-drug-outcome	Non–small-cell lung cancer	> 1,800 patient cases, 150 EGFR variants
My Cancer Genome82	My Cancer Genome is a knowledge base describing the clinical significance of common tumor gene mutations and their relationships to disease and drugs. It includes annotations of gene directed cancer clinical trials27	Tumor variant-disease-drug-outcome	All cancers	> 10,000 active cancer trials annotated with over 400 cancer genes. Detailed reviews for 17 cancer types, 25 cancer genes, over 280 tumor variant-disease relationships
Atlas Genetics Oncology83	Atlas Genetics Oncology is a peer reviewed on-line journal and database in free access on internet devoted to genes, cytogenetics, and clinical entities in cancer, and cancer-prone diseases. It presents concise and updated reviews (cards) or longer texts (deep insights)29	Tumor variant–disease relationships	All cancer diagnoses with a focus on hematologic malignancies	Tumor cytogenetic aberrations in all 24 chromosomes. Not currently up to date in all topics
cBio Cancer Genomics Portal84	This portal contains data from multiple large-scale cancer genomic data sets and allows for explorative analysis	Tumor variant–pathway–disease relationships	19 cancer types	The portal data includes 7,848 tumor samples from 26 cancer studies

Table 2.

Gene–Drug Knowledge Bases to Assist in Clinical Interpretation of Tumor Variants

Knowledge Base	Description	Types of Relationships	Clinical Domains
Pharmacogenomics Knowledge Base (PharmGKB)85	Contains annotations on genetic variants and gene–drug–disease relationships via literature reviews	Gene–drug, germline variant–drug–host response	Somatic and germline pharmacogenomics
DrugBank72	A compilation of detailed drug data with comprehensive drug target information	Gene–drug, drug–pathway	Somatic and germline pharmacogenomics
STITCH86	An interaction network of genes, drugs, and proteins	Gene–drug, gene–chemical	Somatic and germline pharmacogenomics
Therapeutic Target Database (TTD)87	A database of known and explored therapeutic drug targets, the targeted disease, pathway information, and the associated drugs	Gene–drug, drug–pathway, drug–target	Somatic and germline pharmacogenomics

Challenges with respect to evolving clinical knowledge.

An active area of study concerns the questions of if, when, and how to notify providers when new clinical evidence emerges regarding clinically actionable genomic alterations in tumor specimens previously reported. The current standard of care in laboratory medicine is to report the most up-to-date knowledge regarding the clinical significance of the test result at the time the initial report is created. There are no standards for reanalyzing the data or contacting providers when new knowledge or analysis algorithms become available. Ethical guidelines on this matter are being considered for germline studies,⁸⁸ but are lacking for somatic studies. These issues are especially important when new therapies become available (as standard of care or clinical trials) or when new knowledge emerges regarding variant-specific drug sensitivity or resistance. The dynamic reports described in Formatting the Report for a Clinical Audience may be one way to address this issue but require the clinician to refresh the report at the time of transitions in treatment. The next section describes other opportunities for more active forms of clinical decision support to potentially address this issue. However, guidelines and realistic workflow strategies are still needed.

Clinical Decision Support Tools

Reporting results comprises just one way to provide clinicians with decision support regarding potentially clinically actionable results. This type of “passive” clinical decision support requires the clinician to review the report and supporting evidence at the point of making a clinical decision. In contrast, “active” forms of clinical decision support, such as alerts and notifications, are interruptive at the point of finalizing clinical decisions. Active decision support includes alerts for possible contraindications to a selected therapy at the point of provider order entry. For example, if the clinician is attempting to prescribe erlotinib for a patient with EGFR-mutant lung cancer, a decision support system could alert the clinician if the tumor has one of the variants that is known to confer primary or secondary resistance to erlotinib, thereby sparing the patient from a potentially ineffective therapy. The advantage of this approach is the ability to update clinical evidence at the point of treatment action, a major benefit over static reporting strategies. Other opportunities exist to filter and prioritize the list of potential therapies within a chemotherapy order management system at the point of plan selection, taking into account multiple clinical features including the tumor genome. However, such tools are under development for only a small subset of genomic biomarkers.

In conclusion, this science of computational cancer medicine is still in its infancy. With the decreasing cost of tumor gene alteration testing and the increasing number of targeted therapeutics, we expect the methods for analysis and interpretation of cancer genomes to continue to evolve to meet the needs of patient-centered clinical decision making. By successfully bridging research-oriented genomics analysis tools with clinically focused interpretation methods, oncologists may be able to incorporate vast amounts of genomic information into individual patient care. Although it is as yet unknown if this amount of data will result in improvements in clinical outcomes, it is now conceivable to harness the weight of these data and rationally design clinical trials to answer this question across cancer types.

There remains a pressing need to continue to develop knowledge bases, best practices, tools, and validation experiments for the clinical analysis and interpretation of the cancer genome. This effort will require significant and longitudinal collaborations between medical oncologists, molecular pathologists, computational biologists, and biomedical informaticians working together to implement prospective large-scale sequencing. As these tools are developed in conjunction with prospective assessments of the impact cancer genomics may have on patient care, robust analytic and interpretive frameworks should emerge to facilitate clinical cancer genomics throughout the field and across care centers.

In addition, as such analytic and interpretation platforms become more refined, it will be imperative to ensure that these tools and knowledge bases are open source and web based so that they may be accessed by physicians and patients regardless of the care center context. This may require increased regulation of these tools by the US Food and Drug Administration, especially for interpretation algorithms, so that genomic data are not used irresponsibly. Finally, the ethical ramifications of the large amount of data being revealed by these analytic and interpretation algorithms cannot be overlooked; collaborations between ethicists and oncologists armed with this data are necessary and reinforce the need for national guidelines that help guide clinicians on how to safely use information that may have an impact on the patient and his or her family.

GLOSSARY TERMS

Breadth of coverage:

The amount of genomic territory sequenced in a given testing modality.

Copy number alteration:

A structural variation in the genome with an increased (amplification) or decreased (deletion) number of copies of a gene or region.

Depth of coverage:

The number of times a particular place in the genome (eg, base, exon, or region) has been sequenced, often to ensure data accuracy and sensitivity.

Exome:

Part of the genome formed by genes that code for proteins and other functional gene products (known as exons).

Genome:

The complete set of genetic material.

Genome testing modality:

The technology or platform that is used to translate biologic tissue into genetic or molecular data, including Sanger sequencing, mass spectrometric genotyping, allele-specific polymerase chain reaction–based technologies, and massively parallel sequencing, among other options.

Germline alterations:

Alterations in a gene present at conception that are incorporated into every cell of an individual.

Hotspots:

Sites in a specific gene known to harbor recurrent alterations that have well-defined biologic and/or clinical significance.

Insertion/deletion (“indel”):

A local net gain or loss of nucleotides (generally, between one and 50 bases) that results in a frameshift event.

Massively parallel sequencing:

A high-throughput approach to sequencing DNA or RNA, also known as next-generation or second-generation sequencing.

Mutation:

A change of one base in a nucleotide sequence that may result in a change in the amino acid sequence.

Rearrangement:

A genomic alteration resulting from a chromosomal breakpoint that leads to a large structural change in or between chromosomes.

Somatic alterations:

Alterations in DNA that occur in an individual after conception; here used to denote alterations that occur uniquely in a tumor cell.

Targeted sequencing:

Sequencing the coding regions or other selected regions from a relatively small subset of genes.

Transcriptome:

The complete expressed product of the entire genome in a particular cell, tissue, or biofluid at a specific point in time.

Variant:

A change in the DNA sequence from the reference genome that may or may not have functional consequences.

Variant calling:

The method for identifying variants in a tumor sample.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Although all authors completed the disclosure declaration, the following author(s) and/or an author's immediate family member(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships markedwith a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors

Employment or Leadership Position: None Consultant or Advisory Role: Nikhil Wagle, Foundation Medicine (C) Stock Ownership: Nikhil Wagle, Foundation Medicine Honoraria: None Research Funding: None Expert Testimony: None Other Remuneration: None

AUTHOR CONTRIBUTIONS

Manuscript writing: All authors

Final approval of manuscript: All authors