PANDA: pathway and annotation explorer for visualizing and interpreting gene-centric data

What is PANDA?

We have developed a genomics results reporting tool called PANDA (Pathway AND Annotation) explorer. PANDA enables the visualization of multiple annotations in the context of pathways. Annotations in this context are a broad term that refers to various genomics features and information. Annotations can be one of three modalities. First, they can be extracted from a biospecimens such as SNVs, CNVs, structural variants, or gene expression. Second, annotations can be information extracted from public or internal data sources (Abecasis et al., 2012; Sherry et al., 2001) such as frequency of variants, known associations between gene, diseases, and drug-gene relationship. Finally, annotations could also be predictions by software applications (reviewed in Wu & Jiang, 2013) such as the deleterious nature of a mutation. The number of possible annotations is innumerable, and each type may require its own details to make it useful. This diversity makes it very difficult to model all annotations under a traditional method. For instance, genomic information of clinical relevance can include the number of variants in the gene, the position and frequency of these variants in the general population, and the nature of the variant (deleterious or benign). However, these variant-centric annotations can be difficult to reconcile with gene-centric annotations such as the expression level of a gene, the methylation status of a gene profile of a gene, the druggability of a gene, etc. Instead, PANDA summarizes annotations at the gene level and uses an innovative icon-based representation to display these features on biological pathways maps. The use of icons reduces the cluttering of the display, facilitating the visual integration and interpretation of annotations with pathways information. For instance, the relationship between a mutated oncogene that results in a downstream gene becoming up regulated can easily be spotted.

PANDA is also designed to help bioinformaticians deliver gene-centric results in a form more readily interpretable by researchers and clinicians. The application assumes that the inputs have undergone quality and disease-relevance filtering so that it only displays relevant information. PANDA includes an authentication and access-control mechanism to facilitate sharing of dataset between team members and collaborators. Altogether, the tool allows users to select and visualize pathways, toggle annotations views, perform enrichment analysis, and authorize sharing of data with collaborators.

PANDA differs from other pathway visualization tools because users can upload and visualize any number of annotations with any type of content. Icons can be selected to represent a type of annotation (e.g., mutation, expression, etc.) in order to provide visual cues as to what the data represents which is helpful when there are multiple annotations loaded. PANDA also links genes to GeneCards and pathway-level enrichment analysis can be performed on the fly. For convenience, PANDA is pre-loaded with several commonly used annotation sources. This includes 19,777 gene-drug relationships from the Drug Gene Interaction Database (DGIDB) (Griffith et al., 2013), 5,002 gene entries from MalaCards (Rappaport et al., 2013), 3,945 genes from Online Mendelian Inheritance in Man (OMIM, http://omim.org/), 3,243 genes from Human Phenotype Ontology (HPO) (Kohler et al., 2014), and 56 genes from the Pharmacogenomics Knowledgebase (PharmGKB) (Whirl-Carrillo et al., 2012). Details and the code used to generate these annotations are available on our GitHub site.

Input file format for annotation and annotation sets

Since PANDA is not exclusive to any single technology platform (e.g., proteomics, gene expression, DNA sequencing, etc.), there are an immense number of ways that data and annotations can be represented. The data sources are often large and highly complex, thus requiring bioinformaticians to preprocess, annotate, and filter data using appropriate methods for the study. As such, we have designed PANDA to accept as input a simple tab-delimited file format. Each file requires a gene symbol in the first column, followed by one or more annotation field(s) that will be displayed to the user in a later step. Adding a “#” sign to the header line ensures that they table header is transferred to the pathway level view. Each uploaded file is one source of annotation and is assigned a single icon to represent the underlying data.

Operation

Login

Figure 1 and the following text describe how to navigate through the application. PANDA includes an authentication mechanism equipped with logging in and verifying passwords. Users are required to register (which is free) before logging into PANDA. This authentication mechanism is coupled to the access control and data sharing feature (see below), so a user must be registered in the system before data can be shared with them.

Annotation counts and enrichment analysis

Once the annotation files are uploaded, the main page displays the number of annotated genes per pathway per annotation. These columns, like any other column, can be sorted to quickly view the pathways with the largest or smallest number of annotated genes. To identify pathways that have more genes annotated than would be expected by chance, enrichment analysis can be performed on each uploaded annotation dataset, using the function located under the ‘Enrichment’ tab. The end result is an additional column in the main table containing the corresponding p-value from a Fisher’s Exact test.

Use case 1: quickly comparing one’s own data to a published set

Papers are commonly presenting large datasets as supplemental materials. An example is a paper we published previously in a study of pancreas cancer (Murphy et al., 2013). Supplemental table 2 of that paper shows the insertions and deletions per sample. Now let’s say a user is interested in finding out if any of those mutated genes are known to OMIM, HPO terms, and subsets of their own data. Once the table is downloaded, users simple need to rearrange the “Gene” column to be the first, renaming the column header from “Gene” to “#Gene,” choosing which other columns they would like to persist, and saving as a tab-delimited format. Once loaded, any genes in common between the user’s dataset and from the supplemental material will now be represented with two icons next to those genes, instead of just one.

Use case 2: presenting and sharing information in a clinical research setting

PANDA has proven valuable in the genomic oncology clinic at our institution. In the Individualized Medicine clinic, patients with advanced malignancies with limited standard treatment options can undergo next generation sequencing of their tumor in an attempt to find targetable variants. The level of sequencing can vary from limited gene panels of 50–200+ genes at one extreme, up to combined whole genome sequencing (WGS), RNA sequencing (RNA-Seq), and array CGH (aCGH) at the other. Once the sequencing is completed, the data is filtered through various bioinformatics pipelines and discussed at a Genomic Tumor Board (GTB). Only significant results from copy number assessment, differentially expressed genes, or relevant annotations are provided as input into PANDA so that the clinicians are not overwhelmed by trying to view all the raw data from different experiments simultaneously. The GTB then discusses the relevance of the various targets and attempts to create a treatment plan for the patient.

As an example, PANDA was used in evaluating the genome and transcriptome of a 55-year-old Caucasian male with metastatic sarcomatoid RCC with pulmonary metastases. Imaging demonstrated a large renal mass, retroperitoneal lymphadenopathy, and pulmonary masses. A biopsy of the kidney lesion established the histology. The patient elected to undergo genomic analysis of the tumor with WGS (tumor and germline), RNA-Seq, and aCGH. The aCGH showed amplification of YAP1, while WGS demonstrated P287T variant of CCND1 with evidence of possible allele specific expression by RNA-Seq. Figure 2 shows how the data are displayed for all of the assays performed on this patient. This combination of abnormalities was particularly intriguing as YAP1 amplification has been shown to drive CCND1 transcription (Mizuno et al., 2012) and the P287T variant is hypothesized to inhibit polyubiquitination of CCND1, thereby inhibiting its degradation and promoting tumorigenesis (Moreno-Bueno et al., 2003). CCND1 activity is therapeutically targetable by inhibition of CDK4/6 (Musgrove et al., 2011), a target for which multiple agents are currently in clinical trials. The tumor also had multiple other potentially relevant variants including amplification of BIRC3, point mutations in ATM, and a splice variant of TP53. However, the presence of two variants both amplifying the same pathway formed the most compelling narrative for a driver pathway in this tumor, ultimately forming the basis for our treatment recommendation to start a CDK4/6 inhibitor.