CROssBARv2 is a heterogeneous knowledge graph designed to integrate diverse biological and biomedical data from 34 databases, encompassing various entity types such as genes, proteins, and diseases, along with their relationships. This integration enables comprehensive querying and analysis of complex biological systems, supporting downstream molecular biology tasks. The deployment of CROssBARv2 presents challenges due to its scale and complexity, which are addressed through a robust technological framework.

To ensure scalability and efficiency, the deployment leverages Docker with a microservices architecture, isolating faults and enabling horizontal scaling by increasing replicas as needed. Traefik functions as a reverse proxy, managing SSL certificates, load balancing, and rate limiting to maintain stable performance under varying loads. GraphQL is employed for efficient data retrieval, allowing users to fetch only necessary information in a single query, thereby reducing bandwidth usage and enhancing system performance.

The continuous integration and deployment (CI/CD) pipeline is automated using GitHub Actions, which triggers Docker image builds upon each repository push. Watchtower updates containers incrementally across replicas, ensuring smooth scaling and efficient resource management during updates. This combination of technologies—Docker, Traefik, GraphQL, Watchtower, and GitHub Actions—positions CROssBARv2 as a scalable, resilient, and cost-effective solution for various biomedical applications.

CROssBARv2's configurability through environment variables makes it adaptable to diverse use cases, supporting flexible data management for researchers globally. The deployment strategy and technological approach are applicable to other bioinformatic services, facilitating the development, implementation, and manipulation of large-scale data resources efficiently.

The NASA Open Science Data Repository (OSDR) enables open access to space-related data from experiments and missions that investigate biological and health responses of terrestrial life to spaceflight. These data include both omics data (transcriptomics, genomics, proteomics, etc.) from NASA GeneLab as well as non-omics data (physiologic, phenotypic, behavioral data, etc.) from the Ames Life Sciences Data Archive. To fulfill the FAIR principles of open science (Findability, Accessibility, Interoperability, and Reusability), OSDR works with the scientific community via Analysis Working Groups (AWGs) to help define both metadata standards and develop consensus data processing pipelines which are used to generate and publish processed data. For complex datatypes, such as omics data, providing processed data in a standardized, easy-to-use format is vital for facilitating reuse and enabling comparisons across datasets without requiring an extensive computational background or access to compute resources beyond a standard laptop.
Here we present the framework we have developed to design standardized data processing pipelines, implement them as executable workflows, and generate processed data. All of GeneLab’s consensus processing pipelines and workflows are made publicly available on the GeneLab Data Processing GitHub repository, and GeneLab processed data products are available through the OSDR data repository. To-date, GeneLab-processed data from hundreds of space-relevant studies have been reused for multi-modal and multi-hierarchical investigations across experiments and datasets. This has led to new insights and scientific publications that extend beyond the initial research, thereby enriching our understanding of molecular-scale biological responses to the space environment.

We present PRISM (Perturbation Response Integration of Single-cell Measurements), a comprehensive collection of over 70 harmonized public single-cell datasets comprising more than 10 million cells with standardized metadata annotations. This resource integrates diverse perturbation experiments across multiple organisms, cell types, and CRISPR modalities. The PRISM collection addresses a critical challenge in computational biology: the lack of standardized, large-scale perturbation data necessary for building predictive models of cellular behavior. We implemented rigorous preprocessing protocols to ensure consistency across datasets, including standardization of gene expression values, harmonization of metadata, and quality control filtering. All datasets are provided in .h5ad format with uniform annotation structures to facilitate seamless integration. Leveraging this curated resource, we developed a preliminary AI model using flow matching techniques and gene embeddings (GenePT) to predict cellular responses to genetic perturbations. Our model architecture employs neural ordinary differential equations to simulate the trajectory from unperturbed to perturbed states, conditioned on gene embeddings that capture the biological context of the perturbation target. This work represents a significant step toward building a virtual cell - a comprehensive computational model capable of simulating cellular responses to arbitrary perturbations. By integrating data across diverse experimental conditions and developing predictive models, we establish a foundation for in silico prediction of cellular behavior. The PRISM dataset and model code are publicly available to foster community-driven advancement in this field, with potential applications in understanding gene regulatory networks, drug development, and personalized medicine.

To enhance bioinformatics research efficiency by tackling solution discovery, software usage, and database query challenges, we developed BioinfoGPT, an LLM-based multi-agent system. Key methods include a multi-agent architecture and an automated knowledge base workflow: LLMs, guided by prompt learning, processed 80,316 papers classified via EDAM, creating an auto-updating structured database of 46,427 solutions (Feb 2025). An LLM-friendly software doc library and the BSDQA benchmark (multi-difficulty, hallucination distractors) were also built, alongside integrated NCBI/BLAST query tools. Results show BioinfoGPT achieved 0.95 overall accuracy on BSDQA (advanced 0.91), significantly outperforming GPT-4o (0.90/0.86). It scored perfectly (1.0) on gene SNP association and name conversion tasks, demonstrating effective database interaction. This proves our approach combining multi-agent design and auto-updating knowledge enhances LLM performance and reliability for bioinformatics, lowering analysis barriers.

The rapid advancement of whole-genome sequencing (WGS) has transformed cancer genomics, enabling high-resolution profiling of somatic alterations and new insights into tumour evolution. Studying cancer through an evolutionary lens is essential for understanding tumour heterogeneity, subclonal dynamics, and therapeutic resistance. While pipelines for detecting somatic variants are now well-established, tools for inferring tumour evolution from WGS data remain limited.
We present nf-core/tumourevo, an open-source, modular, and fully reproducible pipeline for analysing tumour evolution from next-generation sequencing data. Developed within the nf-core ecosystem using Nextflow, the pipeline integrates leading tools for variant annotation (VEP), driver gene identification, copy number alteration (CNA) quality control (CNAqc), subclonal deconvolution (PyClone-VI, VIBER, MOBSTER), and mutational signature analysis (SigProfiler, SparseSignatures).
nf-core/tumourevo supports both single-sample and multi-sample datasets, enabling analysis of tumour evolution in spatially or temporally resolved cohorts. Its containerized and portable design ensures reproducibility across high-performance computing and cloud environments, making it suitable for both research and clinical applications.
We applied the pipeline to a longitudinal WGS dataset of colorectal cancer, successfully reconstructing clonal architectures and tracking subclonal dynamics over time. We also benchmarked its accuracy using a Simulated Cohort of Universal Tumours (SCOUT), created with our in-house cancer evolution simulator.
By automating complex evolutionary analyses within a reproducible framework, nf-core/tumourevo empowers researchers to explore tumour evolution at scale, advancing biological discovery and supporting precision oncology efforts.