St. Jude KIDS26 BioHackathon Challenges

Title: AI-Assisted Chat Interface for Clinical Data  

Submitter: Sri Sivani Charan Yalamanchi (Biostatistics)  

Summary: Researchers currently rely on data analysts to retrieve information from complex clinical datasets, which introduces delays and limits iterative exploration of data. This project proposes a conversational AI interface using retrieval-augmented generation and Text-to-SQL to enable natural language querying, allowing users to directly access and interact with structured datasets in real time.  

Benefit: The solution enables self-service access to complex datasets, significantly reducing reliance on intermediaries and shortening query turnaround time. It also enhances research productivity by enabling faster hypothesis testing, deeper data exploration, and more efficient use of institutional data resources.  

Tools: RAG, Text-to-SQL, LLMs, Python, PostgreSQL  

Test Data: MIMIC III/IV datasets  

Any PHI, sensitive information, or otherwise confidential data use: No  

Title: AI-Enabled Structuring of ARIA Guide Clinical Recommendations  

Submitter: Monica Key (Global Pediatric Medicine)  

Summary: Clinical recommendations in ARIA are currently documented in narrative formats and manually converted into structured templates, creating inefficiencies and inconsistencies. This project develops an AI-assisted pipeline to extract key clinical elements and transform them into standardized, structured formats for downstream use in analytics, tracking, and digital tools.  

Benefit: Automating this process reduces time-intensive manual work and improves consistency and traceability across guideline versions. It also supports scalable implementation of “living guidelines” and provides a reusable framework for structuring clinical knowledge across programs.  

Tools: AI/NLP tools  

Test Data: ARIA Word documents  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: AI/ML-Driven Sequencing Run Planner for NGS Core Operations  

Submitter: Daniel Darnell (Developmental Neurobiology)  

Summary: Sequencing run planning currently relies on manual expertise, often leading to inefficient use of sequencing platforms and variability in outcomes. This project develops an AI/ML-driven system that leverages historical quality metrics and run data to recommend optimal sequencing configurations and predict potential risks.  

Benefit: The system improves sequencing efficiency by reducing failed or suboptimal runs and maximizing utilization of sequencing capacity. It also standardizes planning workflows, lowers operational costs, and increases throughput in core sequencing operations.  

Tools: Machine learning models  

Test Data: Sequencing QC and run metrics  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: AnTelOpe: An LLM-Powered Web Platform for Nextflow Pipeline Orchestration and Monitoring  

Submitter: Haidong Yi (Center for Applied Bioinformatics)  

Summary: Bioinformatics pipelines require significant expertise to configure, submit, and troubleshoot, limiting accessibility for many researchers. This project integrates an LLM-based conversational assistant into a pipeline platform to support natural language-driven configuration, monitoring, and interpretation of workflows.  

Benefit: The approach reduces technical barriers for non-expert users and significantly accelerates pipeline execution and troubleshooting. It enables more efficient use of computational infrastructure by simplifying interaction with complex systems and improving overall user experience.  

Tools: LLM chatbot, Nextflow  

Test Data: Pipeline logs and workflow data  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: AutoCAB: Turning Everyday Scientific Workflows into Reusable AI Agent Skills  

Submitter: Niu Beifang (Center for Applied Bioinformatics)  

Summary: Many scientific and bioinformatics workflows are repeatedly performed but remain undocumented or inconsistently implemented. This project identifies recurring workflow patterns and converts them into reusable AI-driven “skills” that can automate common tasks across projects.  

Benefit: The system reduces redundant manual effort and improves productivity across teams. It also enables standardization, knowledge sharing, and reproducibility by transforming tacit workflows into reusable, structured capabilities.  

Tools: AI agents  

Test Data: Workflow traces  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Cell Identity in Spatial OMICs Imaging Platforms: Cluster-Based vs Cell-Based Annotation  

Submitter: Maycon Marção (Developmental Neurobiology)  

Summary: Spatial transcriptomics relies on different strategies for cell type annotation, yet the relationship between cluster-based and rule-based approaches is not well understood. This project compares these methods to evaluate their strengths and limitations in interpreting spatial omics data.   

Benefit: The results improve confidence in cell-type annotations and biological interpretation. They also support development of best practices and standards for spatial omics analysis.   

Tools: Spatial transcriptomics analysis  

Test Data: Imaging-based transcriptomics datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: CLEAR: Clinician-Led, Easy-to-read, Accessible Reaffirmed Instructions  

Submitter: Maria Lopez Garcia (Hospitalist Program)  

Summary: Patients are often given complex instructions at the end of clinical encounters when they are least able to absorb and retain information. This project focuses on designing structured, easy-to-understand instructions that patients can revisit at home, improving clarity and usability.  

Benefit: Clearer communication improves patient understanding and adherence to treatment plans, particularly in high-risk care contexts. It also reduces the likelihood of errors and enhances overall patient outcomes and safety.  

Tools: Not specified  

Test Data: Clinical instructions  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Counting every cell in the human brain: 3D localization of cell nuclei in a human brain  

Submitter: Caitlin Freeman (Imaging Sciences)  

Summary: Detecting and localizing cell nuclei in large volumetric brain imaging datasets is computationally challenging at scale. This project develops efficient and scalable segmentation methods to process large 3D imaging data and identify cellular structures.  

Benefit: The work supports large-scale brain atlas development and advances understanding of brain structure at the cellular level. It also improves computational efficiency for analyzing extremely large imaging datasets.  

Tools: Machine learning segmentation  

Test Data: Light-sheet microscopy datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: CureCheck: An R Shiny Application for Evaluating Cure Model Appropriateness  

Submitter: Durbadal Ghosh (Biostatistics)  

Summary: Cure models in survival analysis are often applied without carefully evaluating underlying assumptions, which can lead to misleading results. This project develops a tool to assess when cure models are appropriate and guide their correct application.  

Benefit: The tool strengthens statistical rigor in clinical research and supports more reliable interpretation of survival outcomes. It also helps avoid incorrect conclusions and improves methodological decision-making.  

Tools: R Shiny  

Test Data: Survival datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Establishing a Sample-Based Quantum Diagonalization (SQD) Workflow for Iron-Sulfur Cluster Electronic Structure  

Submitter: Amandeep Bhatia (Structural Biology)  

Summary: Iron–sulfur clusters are biologically critical but difficult to model due to their complex electronic properties and interactions. This project develops a sample-based quantum diagonalization workflow to enable more accurate representations of these systems.  

Benefit: The approach improves computational modeling of key biochemical systems, enabling better understanding of electron transfer and catalysis. It also contributes to advances in quantum-enabled modeling techniques for biological systems.  

Tools: Quantum modeling approaches  

Test Data: Fe-S model systems  

Any PHI, sensitive information, or otherwise confidential data use: No  

Title: FAIR Harmonization and Automated LabKey Integration of Legacy Sequencing Metadata  

Submitter: Cody Ramirez (Developmental Neurobiology)  

Summary: Legacy sequencing datasets often contain inconsistent, incomplete, and fragmented metadata across studies and platforms. This project develops automated processes to harmonize, validate, and standardize metadata before integration into centralized systems such as LabKey.  

Benefit: The solution improves data quality, reproducibility, and interoperability across datasets. It also enables efficient reuse of historical data and supports large-scale data integration efforts.  

Tools: Data harmonization pipelines  

Test Data: Metadata files  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Genomic Scars Pipeline: Assessment and Visualization Shiny  

Submitter: Evan Savage  

Summary: Existing tools do not provide an integrated platform for assessing and visualizing genomic scars, particularly in pediatric cancer datasets. This project develops a Shiny-based application to enable analysis and interactive exploration of genomic instability features.  

Benefit: The platform improves understanding of genomic signatures associated with therapeutic response and disease progression. It also supports translational research by enabling researchers to interpret complex genomic data more effectively.  

Tools: R Shiny  

Test Data: WGS datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: GRIN3D: From Linear Sequence to 3D Protein Hotspots  

Submitter: Abdelrahman Elsayed  

Summary: Traditional genomic analysis focuses on linear sequence data and does not capture the three-dimensional structure of proteins where functional effects occur. This project extends hotspot detection into 3D protein space to better represent biological impact.  

Benefit: It improves understanding of how mutations affect protein structure and function. It also provides deeper insights into disease mechanisms and supports development of more accurate predictive models.  

Tools: Statistical frameworks  

Test Data: Genomic/protein datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Integrating Molecular and Clinical Trial Data to Predict AML Patient Outcomes for Novel Therapies  

Submitter: Stanley Pounds (Biostatistics)  

Summary: Translating molecular responses observed in laboratory models to clinical outcomes remains a challenge in oncology. This project integrates molecular data with clinical trial datasets to predict survival outcomes for novel therapies in AML.   

Benefit: It enables prioritization of promising therapies for further investigation. It also accelerates translational research by linking experimental findings to clinical impact.   

Tools: Statistical modeling  

Test Data: GEO and clinical trial datasets  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Multimodal Sleep Disordered Breathing (SDB) Event Classification  

Submitter: Miguel Navarrete  

Summary: Detecting sleep apnea events from home monitoring systems is challenging due to noise and variability in physiological signals. This project develops machine learning models to classify events from multimodal time-series data.  

Benefit: The system improves diagnostic accuracy and enables scalable, home-based monitoring solutions. It also enhances early detection and management of sleep disorders.  

Tools: Machine learning models  

Test Data: Physiological signals  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: Resource-Aware Multi-Project Workflow Orchestrator for Scalable Single-Cell Analysis  

Submitter: Antonia Chroni  

Summary: Current bioinformatics workflows are not designed to efficiently handle multiple concurrent projects, leading to suboptimal resource usage. This project develops a system to dynamically allocate compute resources and manage workflows across projects.  

Benefit: It improves efficiency and scalability in high-performance computing environments. It reduces resource waste and supports faster execution of large-scale analyses.  

Tools: HPC workflow systems  

Test Data: Pipeline workloads  

Any PHI, sensitive information, or otherwise confidential data use: Not specified  

Title: SCOGS-Scribe: Automated Severity Grading for Sickle Cell Disease from Clinical Notes  

Submitter: Joe Wardell  

Summary: Sickle cell disease severity is complex and inconsistently captured in clinical documentation. This project uses natural language processing to extract structured severity scores from clinical notes.  

Benefit: It enables consistent and reproducible severity grading across datasets. It also supports clinical research, trial design, and improved patient management.  

Tools: NLP  

Test Data: Clinical notes  

Any PHI, sensitive information, or otherwise confidential data use: Likely yes  

Title: Where Healing Happens: Mapping Mental Health in Childhood Cancer  

Submitter: Sylwia Feibelman  

Summary: Mental health outcomes in pediatric cancer patients may be influenced by environmental and socio-economic factors that are not typically considered in clinical analyses. This project integrates deidentified patient data with geospatial datasets to explore these relationships and generate insights.  

Benefit: The approach identifies geographic and environmental factors that impact patient outcomes and supports targeted interventions. It also advances understanding of social determinants of health in pediatric oncology.  

Tools: GIS, statistical analysis  

Test Data: Deidentified patient + environmental data  

Any PHI, sensitive information, or otherwise confidential data use: No