How to Compare Gene Signatures on Polly

Shruti Malavade, Shrushti Joshi
May 17, 2023
How to Compare Gene Signatures on Polly

Gene Signature comparisons with available datasets have proven to be a powerful technique utilized by biopharma R&D teams for drug discovery, biomarker identification, development, and personalized medicine.

This technique allows researchers to analyze the expression levels of large numbers of genes in samples from individuals with a particular condition or disease and compare it to a conserved cluster of genes whose expression levels are most strongly associated.

This gene signature can then be used to search public databases of gene expression data for other drugs or compounds that can revert the disease signature, indicating a potential therapeutic effect.

However, extracting associated signatures from public databases can be challenging due to various processing pipelines, syntaxes, schemas, and metadata annotations used at the source. We address these challenges through Polly’s RNA-Seq Omixatlas.

This blog discusses how users can compare signatures using Polly's RNA-Seq OmixAtlas.

What is Polly?

Polly is a biomedical data platform for life sciences R&D, primarily delivering bulk & single-cell RNA-seq data, along with 24 other data types. It delivers 155 TB of FAIR and ML-ready biomedical data from ~30 different public and proprietary sources to customers. Polly’s RNA-Seq OmixAtlas (OA) contains curated RNA seq datasets collected from Gene Expression Omnibus (GEO). This richly curated resource provides a good base for researchers looking to find datasets with similar transcriptional profiles to their gene sets of interest.
All datasets on Polly are:

  • Consistently Processed
    End-to-end data processing (identifier mapping, QC, normalization, and alignment) is orchestrated through the Kallisto pipeline.  Consistent processing on the entire Atlas allows samples to be reliably combined into cohorts and used to develop RNA-Seq signatures.
  • Enriched with Metadata
    All datasets are enriched with over 21 searchable metadata fields (disease, gene, tissue, drug, control, etc.) at the dataset, sample, and feature levels. This means that users can quickly run SQL queries to find datasets with normal to-disease comparisons and define cohorts of their choice.

Our Approach

How to Compare Gene Signatures on Polly

Generate Query Signature(S)

The first step to compare Gene Signatures is to create a query wherein the gene of interest can be searched against a dataset to identify a closely associated gene cluster. To generate a query signature, the following steps are required:

  1. Define your query: Define the biological process or condition of interest for which one wants to generate a gene signature.
  2. Choose a dataset: Select a dataset containing gene expression data for your query’s samples.
  3. Pre-processing the data: This involves normalizing the data, filtering out low-expression genes, and correcting for batch effects.
  4. Identify differentially expressed genes: Statistical methods such as limma or DESeq2 are used to identify differentially expressed genes between your query and control groups.
  5. Construct the gene signature: Combine the list of differentially expressed genes into a gene signature using a method such as a gene set enrichment analysis (GSEA), principal component analysis (PCA), or support vector machine (SVM).
  6. Validate the gene signature using independent datasets or experimental validations to confirm its relevance to your query.

Or Polly experts can be contacted that will work with your scientists to customize these steps as needed and capture transcriptome profiles and generate queries (gene signature vectors) that will run on Polly’s signature database. The query will consist of gene clusters that were significantly differentially expressed in the experiment with Log Fold Change, p-values, and adjusted p-values

Example of a query: Given an input of gene set and Log Fold Change values, search for all datasets that show maximum cosine similarity scores with the input genes and their differential expression results.

Creating a Signature Database Derived from Data on Polly

  • Experiment designs of all RNA-Seq datasets on Polly are evaluated. Datasets containing control and perturbation samples are then extracted from this collection.
  • A differential gene expression analysis is performed on these cohorts of control and perturbed samples.
  • The resulting statistical computation includes a distinct ID for the Differential Comparison, Gene Names, and their values for Log Fold Change, p-Value, and adjusted p Values.
  • These results and metadata, such as perturbations, controls, disease, drug, or genotype, are indexed on Polly in query-able .gct files.
  • Simultaneously a database of gene signatures vectors is created based on your choice of thresholds for Log Fold Change and adjusted p-value cut-offs.

Identifying Datasets Similar to the Query Signature

This signature database can now be queried to identify datasets with similar transcriptional profiles to the Query Signature. For instance, users can run complex SQL queries to identify:

  • Datasets where diseased samples are compared to normal and are similar to the query signature.
  • Datasets where a particular disease is treated with some drug and shows a reverse profile to the query signature.
  • Datasets where genes are differentially expressed in cancer cells compared to normal cells, which drugs can target.

Ranking the Output

  • Our experts work with your team to identify your preferred method for finding similar or dissimilar transcriptome profiles from the database and rank these.  We can employ standard scores in literature such as the Jaccard index, cosine similarity, concordance/discordance ratio, etc.
  • We also created a random query gene signature with the same number of differentially expressed genes and obtained the distribution of similarity scores to serve as a background distribution. This helps identify significant similarity scores for the given query signature, which can be downloaded in Excel or copied in .txt or .doc.

Case Study: Predicting Synergistic Drug Combination for COVID-19


We used signature reversal and multivariate gene expression signatures to identify potential drug combinations for COVID-19. To do this, publicly available transcriptomics data from COVID-19 studies and drug signatures from LINCS were compiled, processed, and curated. All datasets were ingested through Polly's proprietary curation pipeline, enriched with ontology-backed metadata, and engineered to a query-able .gct format.

Predicting Synergistic Drug Combination for COVID-19 | Elucidata


  • Thirty-seven reference drug candidates based on the similarity between drugs and disease profiles were identified on Polly. Drugs with low drug-disease similarity across most disease profiles were shortlisted.
  • Drug combinations were evaluated based on similarity to reference drugs and disease signature reversal. Twenty-eight combinations with low reference drug similarity and high disease signature reversal were prioritized.
Want to perform gene signature comparisons effectively? Talk to us!

Request Demo