Prioritization Scores
PandaOmics employs a sophisticated scoring system to prioritize therapeutic targets based on diverse data sources, including multimodal omics datasets, text-based evidence, financial metrics, and expert insights. High score values indicate a stronger association between a gene and the disease of interest, greater druggability, or higher commercial potential, depending on the specific score. Users can view the ranked gene list in the Target ID interface, presented as a dashboard with a heatmap and customizable filters.
Score Categories
PandaOmics organizes its prioritization scores into distinct categories, each leveraging different data types and/or methodologies to evaluate potential therapeutic targets:
-
assess the biological relevance of a gene to a specific disease using multi-omics data, such as gene expression, proteomics, and methylation profiles. These scores are derived from bioinformatics analyses and ML models that integrate data from public and user-uploaded datasets (accessible only to the uploader). -
evaluate the relevance of a gene based on its mentions in scientific literature, patents, grants, and clinical trials. These scores leverage AI-based text mining and natural language processing to quantify attention and trends. -
estimate the commercial viability of a target based on funding trends, extracted from grant applications. -
scores gauge a gene's prominence in high-impact research based on expert publications and journal influence. -
LLM (Large Language Models)scores leverage AI to analyze diverse biomedical text, including scientific literature, patents, and clinical trial reports to evaluate a target’s disease relevance, market potential, and druggability.
Configuring and Using Scores
By default, PandaOmics ranks genes by averaging a set of five core scores optimized for exploring well-studied targets with robust evidence. For researchers investigating less-studied or novel targets, enabling additional scores beyond the core set allows for more refined gene rankings and deeper insights.
Users can also select predefined presets like "Trending Genes" or "Novel Targets (Small Molecules)" to toggle specific combinations of scores and druggability filters tailored to their research focus.
Practical Tips
-
For initial exploration, use the default core scores to identify well-validated targets with strong evidence across categories. - For drug repurposing, explore the Indication Prioritization tab to identify diseases where a target may have high clinical success potential.
- Include scores from the Omics category for emerging or less-studied targets.
- Apply druggability filters to align gene rankings with project goals.
- Use the Knowledge Graph and ChatPandaGPT to dive deeper into the evidence behind top-ranked targets, ensuring robust hypothesis validation.
OMICs
Omics score values represent the probability of a given evidence group indicating the association of a given gene to a given disease. All scores range from 0 to 1, with 0 indicating no evidence, and 1 - the highest degree of evidence
A Network Neighbors score utilizes several graph-based methods applied to the protein-protein interaction network enriched with differentially expressed/methylated genes. A target will be scored higher if there are more network neighbors with significant differences in expression or methylation levels.
This score estimates target relevance from the perspective of indirect genetic evidence. The signal from the GWAS hits and cancer drivers is propagated through the gene interaction network using diffusion algorithms. Genes belonging to the network submodules enriched with genetic variants are scored higher.
This score estimates whether a gene physically interacts with other genes known to be implicated in a specific disease. The score relies on manually curated data and external sources (ClinVar, Open Targets).
Heterogeneous Graph Walk (HeroWalk) is a guided random walk-based approach that is applied to a heterogeneous graph (i.e., a graph containing different types of nodes). The model learns node representations and then finds the gene nodes similar to the reference disease node. First, the "walks" are sampled with a predefined meta-path, i.e., the fixed sequence of node types in a walk, e.g. 'gene'-'disease'-'gene.' The node degree controls the probability of transition between the nodes while sampling. Following that, the AI model learns the representation of each node based on the resulting corpus of walks. The cosine similarity between the specific disease and all available genes produces a ranked list of genes. The top genes from this list are predicted to be promising target hypotheses.
Matrix Factorization score is based on a collaborative filtering algorithm, which is widely used in recommender systems. First, well-known gene-disease associations from the PandaOmics database are converted to a sparse binary matrix.This matrix is then decomposed into two low-rank matrices that consist of latent factors for genes and diseases. The algorithm uses graph regularization based on a fast kNN search to account for the intraclass similarity between the nodes of a similar type. Recomputing the original interaction matrix from latent factors provides the scores for unobserved interactions; thus, gene ranking is obtained.
The score takes into account differential gene expression, protein abundance, or methylation level. Machine learning-based models are used to normalize available omics data within multiple samples from various datasets.
Interactome Community score utilizes several AI graph-based methods applied to the protein-protein interaction network enriched with active drug targets, GWAS hits, and differentially expressed/methylated genes.
This score utilizes a manually curated regulatory network consisting of transcription factors and other genes. A target will be scored higher if its network neighborhood is enriched with transcription factors regulating differentially expressed genes.
This score compares differential expression from the datasets of interest with transcriptomic data on cell lines with perturbation-induced target gene overexpression (LINCS database). A target is scored higher if the differential expression signature is similar to the gene expression changes in perturbed cells.
This score compares differential expression from the datasets of interest with transcriptomic data on cell lines with perturbation-induced target gene knockout (LINCS database). Target is scored lower if the differential expression signature is similar to the gene expression changes in perturbed cells.
This score ranks genes based on the mutation burden using genetic data shown to be associated with current disease/phenotype. Data is compiled from a variety of sources, including Clinvar, GWAS catalog and IntOGen. Higher scores indicate mutation burden correlating with a given disease.
The score combines several approaches to pathway analysis. First, the iPanda algorithm is used to examine the involvement of a given gene in pathway activation patterns in order to get a collection of gene expression datasets of interest (activation/inhibition of each pathway is examined separately). Next, all the pathways from the library are merged into a single network, which is examined from the perspective of signal propagation by a number of methods. The final score indicates how a given gene affects individual pathways activation/inhibition and whether it possesses the ability to affect multiple pathways at once.
The Relevance score is higher for those genes which are known drug targets. It also takes into account the number of clinical trials for corresponding drugs and their phase (late phases are scored higher). The data for the score calculation is obtained from the OpenTargets resource.
Text-based (NLP)
This group of scores is based on the analysis of text sources, including scientific publications, grants, patents, clinical trials. Insilico monitors the biomedical text data and uses advanced NLP and AI-based technology, to analyze the contextualized mentions of entities such as genes, diseases, drugs and KOLs in a variety of data sources. The scores are calculated for each gene in the disease-agnostic manner (total mentions of the gene), and in the context of disease areas of interest. Attention spikes are calculated for target — disease associations only (no disease-agnostic scores).
The Attention score measures the overall attention to the target at all times. PandaOmics calculates the total number of mentions of a gene in various texts across all time periods. Both disease-agnostic and disease-specific mentions are counted. The text corpus used for analysis includes scientific publications, grants, patents, and clinical trials.
The Trend score estimates the disease-specific growth of attention to a particular gene during the 5-year period. It ranges from -1 (no papers were published) to 1 (all papers were published during the last year). The negative values correspond to the decline in publication activity. The exact trend value of a gene is available on the gene page.
This score utilizes AI-based text mining methods to predict the likelihood of the future increase in research interest for the given target and indication. The neural network trained on retrospective publication trends is used to forecast the attention spike probability in the next two years. Text corpus used for analysis includes scientific publications, grants, patents, and clinical trials.
This score is calculated as the weighted average of the Trend and Attention scores. Higher values indicate both attention growth and a high volume of research.
Financial scores
Total grant funding to investigate the given gene at all times. Please note that the distribution of total funding is skewed with the mean value of approx. $8.9 million, but the median value is only $1.7 million. Also, 5% of entries have funding above $30 million.
This score measures the total sum of grant funding per publication to investigate the given gene. The average grant size reflects the scientific community opinion on the relevance of the research of the particular gene or gene-disease association. Entries with zero publications and non-zero funding are scored as zero.
The average grant size for a given target disease association. The total grant funding is divided by the total amount of grants.
Key Opinion Leaders (KOL)
This score measures the average Hirsch index of the researchers who published scientific articles studying given gene-disease association.
This score measures the average impact factor of the journals where a particular gene-disease association is published. The Impact Factor of each journal is weighted to the number of publications mentioning gene-disease association published in this journal.
This score represents the total number of publications devoted to the target gene in journals with an impact factor of more than 10. This corresponds to the top 3% of all scientific journals.
LLM
Measures the evidence supporting the target's role in disease mechanisms, considering experimental validation and robust scientific data. Higher scores reflect solid validation and reduced uncertainties.
Assesses the target's market feasibility, taking into account demand, competitive landscape, and profitability. Higher scores suggest strong market potential and competitive advantages.
Indicates the likelihood of successfully developing therapeutics targeting this protein, based on its structural, functional, and biochemical attributes. Higher scores indicate easier drug development and improved targetability.
Examines the understanding of how the intervention affects pathways, diseases, and resistance mechanisms. Higher scores suggest a well-defined mechanism of action and reduced uncertainties.