The neural network of special architecture, the so-called transformer extracts the relevant information from the list of scientific publications and produces a nice-looking graph that visually links genes, diseases, chemical compounds, and biological processes. The knowledge graph is accessible on the Publications tab of the Gene, Disease, and Gene-Disease Association pages. Explore the knowledge graph to gain insights into the molecular mechanisms of the disease, and biological activities of genes and compounds in the context of the disease, and decide on the most promising drug targets and biomarkers. Click on genes and diseases, rearrange nodes, and use zoom and pan functions to come up with publication-ready images. Hovering the mouse over the graph's edge brings up tooltips with the paper citations supporting the corresponding relation in the graph.

While users always have the option to analyze datasets independently from each other and then combine the expression and pathway results in a Meta-analysis (macro-aggregation), the cross-dataset comparison feature allows merging sample groups together across multiple datasets on earlier steps of analysis (micro-aggregation). This function is available on the Dataset tab in the Data Manager. Batch correction can also be enabled to improve the quality of the results and correct for possible batch effects. The user is also provided with quality metrics and PCA plot, informing whether the batch correction has been performed successfully. Even datasets without paired normal control samples can now be used for the subsequent analysis.

Several tabs on the Disease and Meta-analysis pages have been significantly updated. Access the sample number and omics data types available in the Meta-Analysis directly from the Meta-analysis summary tab. Availability of microarray, RNA-seq, proteomics, methylation, and GWAS data can be found in the omics data summary card. This also applies to the curated disease reports available through the main search. Use filters similar to the Target ID tab to narrow down the list of genes in the Expression Analysis tab. For example, one can quickly obtain a list of kinases with resolved 3D protein structures showing differential expression across various datasets. The publication-ready expression heatmap can now be exported as an image in vector or raster graphics format.

In PandaOmics 3.0 we have significantly improved our named-entity recognition technology. This results in more accurate automatic detection of various biomedical entities like genes, diseases, drugs, etc. Our web crawlers regularly scan the scientific literature, clinical trials, patents, and grant applications to provide you with the most up-to-date information about promising targets and therapeutics. This also ensures AI-based target prioritization scores operate at their maximum performance with the addition of frequent updates to text and clinical trial data.

For this release, our AI experts have compiled and analyzed -omics datasets for most if not all the diseases included in the pipelines of the top 20 pharma companies, globally. This means that PandaOmics users are provided with pre-calculated, out-of-the-box reports for promising diseases and conditions. PandaOmics 3.0 gives immediate access to your prioritized lists of targets and compounds without the need to perform the data analysis yourself. In case of interest in a specific therapeutic condition, the user can rely on the comprehensive PandaOmics database to build new projects. In Version 3.0, more than 1000+ high-quality datasets have been added to the PandaOmics library.

Pre-calculated Target ID and Compound ID are now available for all the diseases. They can be accessed through the Disease page. Just open any disease page and go to the corresponding tabs in the left sidebar or use the links from the Disease Summary tab. The pre-calculated analysis is based on the text data, GWAS, and biological graphs. Manually curated, high-quality Gene expression datasets are also used for omics scores calculation for more than 200 diseases.

Manually curated datasets checkmark is introduced in the Dataset tab on the Disease pages. The datasets flagged with check marks have passed manual quality control of meta-data annotation, study design relevance to the disease, and data quality. Save time on inspecting the datasets manually and select the ones with check marks if available to ensure high quality of the results.

Omics dataset summary tab on the Disease and Meta-analysis pages provides the statistics of the omics data used for subsequent steps of analysis including Target ID and Compound ID.
The total number of genes in the dataset and the portion of the differentially expressed genes is shown. It is also possible to group Comparisons, e.g. by the tissue of origin or other properties of the source samples and datasets. The Comparison groups can be further used to aggregate the results in Target ID.

Target ID: Switching on/off Comparisons is now allowed in the Target ID interface without reassembling and recalculating the entire Meta-analysis. Just use the new feature of the Target ID dashboard to turn on and off Comparisons or Comparison groups, and get the updated ranking immediately. This feature is specifically useful to compare the target ID outputs calculated using public data only versus the proprietary datasets uploaded by the user. Another example is switching between the Target ID results for different tissues.

2 additional AI scores in Target ID were added in the omics section of the target discovery dashboard. Heterogeneous Graph Walk (HeroWalk) is a guided random walk-based approach that is applied to a heterogeneous gene-disease graph. Matrix Factorization is a collaborative filtering algorithm widely used in recommender systems applied to a gene-disease interaction graph. The new scores strengthen the AI component and result in a better performance of the target hypothesis generation engine while preserving the general layout and backward compatibility of the Target ID tool.

More features are added in the Target ID section of the application. Presets allow saving custom states of the filtering options (druggability, tissue-specific expression, list of scores used for ranking, etc.) and applying them in one click to any disease. A collection of default presets e.g. targets for repurposing, and novel targets, is available. A gene of interest can be found using the Gene search box. The definition of the filtering criteria in a druggability traffic light widget is reworked.