ProCyon is a groundbreaking foundation model for modeling, generating, and predicting protein phenotypes across five interrelated knowledge domains: molecular functions, therapeutic mechanisms, disease associations, functional protein domains, and molecular interactions. To train ProCyon, we created ProCyon-Instruct, a dataset of 33 million protein phenotype instructions, representing a comprehensive resource for multiscale protein phenotypes.

Integrating clinical knowledge into AI remains challenging despite standardized medical guidelines and vocabularies. Medical codes, central to healthcare systems, often reflect operational patterns shaped by geographic factors, national policies, insurance frameworks, and physician practices rather than the precise representation of clinical knowledge. This disconnect hampers AI in representing clinical relationships, raising concerns about bias, transparency, and generalizability.

Here, we developed a resource of 67,124 clinical vocabulary embeddings derived from a clinical knowledge graph tailored to electronic health record vocabularies, spanning over 1.3 million edges. Using graph transformer neural networks, we generated clinical vocabulary embeddings that provide a new representation of clinical knowledge unified across seven medical vocabularies. We validated these embeddings through a phenotype risk score analysis involving 4.57 million patients from Clalit Healthcare Services, demonstrating their ability to stratify individuals by survival outcomes. Inter-institutional panels of clinicians evaluated the alignment of embedding with established clinical knowledge across 90 diseases and 3,000 clinical codes, confirming their robustness and transferability.

This resource addresses the gap in integrating clinical vocabularies in AI models and training datasets and supports population and patient modelsin precision medicine.

PrimeKG is a precision medicine-oriented knowledge graph that provides a holistic view of diseases. It integrates 20 high-quality resources to describe 17,080 diseases with 5,050,249 relationships representing ten major biological scales, including disease-associated protein perturbations, biological processes and pathways, anatomical and phenotypic scale, and the entire range of approved and experimental drugs with their therapeutic action.

PrimeKG supports drug-disease prediction by including an abundance of ’indications’, ’contradictions’ and ’off-label use’ edges, which are usually missing in other knowledge graphs. We accompany PrimeKG’s graph structure with text descriptions of clinical guidelines for drugs and diseases to enable multi-modal analyses.

GraphXAI is a resource to systematically evaluate and benchmark the quality of GNN explanations. A key component is a novel and flexible synthetic dataset generator called ShapeGGen that automatically generates a variety of benchmark datasets (e.g., varying graph sizes, degree distributions, homophilic vs. heterophilic graphs) together with ground-truth explanations that address all known pitfalls of explainability methods.

We are developing representation learning techniques for complex time series dataset. In the Raindrop study (ICLR’22), we introduced a graph-guided network for irregularly sampled multivariate time series. The study includes a processed sensor dataset recording daily living activities of individuals.

We present a comprehensive catalog of 10,443,476 adverse event reports (involving 19,193 adverse events and 3,624 drugs) from the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS), collected from January 2013 to September 2020. The new resource can help discover relationships between drugs and safety events, especially in cases of rare events and effects within population subgroups that differ in their risks of specific clinical outcomes and are disproportionately affected by preventable inequities.

We design novel synthetic and real-world social and biological datasets, consisting of underlying base graphs and many labeled subgraphs. These datasets are ready to be used for benchmarking, systematic model evaluation and comparison.

TDC is the first unifying framework to systematically access and evaluate machine learning across the entire range of therapeutics.

At its core, TDC is a collection of AI/ML-ready datasets and learning tasks to serve as a meeting point for domain and ML scientists. TDC also provides an ecosystem of tools, libraries, leaderboards, and community resources, including data functions, strategies for systematic model evaluation, meaningful data splits, data processors, and molecule generation oracles. All datasets and learning tasks are integrated and accessible via an open-source library.

BioSNAP is a collection diverse biomedical networks, inclusing protein-protein interaction networks, single-cell similarity networks, drug-drug interaction networks.

BioSNAP datasets contain metadata on graphs and node features, and can be easily linked to external repositories of biological knowledge.

Graph datasets comprise of critical decisions in criminal justice (if a defendant should be released on bail) and financial lending (if an individual should be given loan) domains. These attributed graphs contain sensitive/protected attributes, which makes them suitable for studying algorithmic fairness.

OGB is a collection of benchmark datasets, data loaders, and evaluators for graph machine learning. Datasets cover a variety of graph machine learning tasks and real-world applications.

The OGB data loaders are fully compatible with popular graph deep learning frameworks, including Pytorch Geometric and DGL. They provide automatic dataset downloading, standardized dataset splits, and unified performance evaluation.

The multimodal cancer network integrates information on chemicals, diseases, molecular functions, genes, and protein.

The dataset has 21 types of biologically meaningful associations (edge types): chemical-chemical, chemical-protein, disease-chemical, disease-disease, disease-function, disease-gene, function-function, gene-gene (split into 6 edge types by interaction type), gene-protein, protein-function, and protein-protein interactions.

The network has 20 K nodes and 3.4 M edges.

The giga-scale biological network is one of the largest networks ever constructed in biology. The network integrates protein and genetic interaction data from more than two thousand species.

The network has 10 M nodes and 2.3 B edges.

The dataset contains protein interactomes from 1,840 species across the tree of life. The dataset contains rich metadata about proteins, including their homology relationships

The dataset also contains metadata about species, including taxonomy of species, phylogenetic reltionships, and ecological information on environments and habitats in which species live.

The polypharmacy network is a highly multi-relational network, consisting of protein-protein interactions, drug-protein targets, and drug-drug interactions encoded by polypharmacy side effects.

The network has 20 K nodes and 5 M edges, which are split into 1 K distinct edge types.

The dataset contains protein-protein interaction networks specific to 107 human tissues, a tissue hierarchy of anatomical relationships between tissues, and tissue-specific gene-function annotations.

The human knowledge network contains interactions between proteins, diseases, biological processes, side effects, and drugs.

The network has 98 K nodes and 8 M edges, which are split into 42 distinct types of biologically relevant molecular interactions.