Omics Data Assistant

Please note: in order to use the Omics Data Assistant on the Platform, a License is required.

Use and Limitations

Summary of Key Disclosures

Category	Summary of Limitation
Accuracy	Responses are based on learned patterns and must be independently verified by the user.
Data Scope	Output is strictly limited to indexed data; complex prompts may yield unexpected results.
User Responsibility	The user is fully responsible for the proper use and interpretation of results.
Non-Deterministic	ODA may provide different responses to the same query (probabilistic nature).
Verification	All outputs, including statistical interpretations, require verification by a qualified professional.
Regulatory	ODA is Research Use Only (RUO), not a medical device, and does not provide medical advice or diagnoses.
Context	Cannot account for biological variables or metadata not explicitly provided.
Experimental	Insights generated (e.g., pathway enrichment) are hypotheses for further validation.
Data Privacy	Strictly prohibited from uploading PII or PHI; all datasets must be de-identified (HIPAA/GDPR compliance).

For a detailed breakdown of these limitations, please refer to the Comprehensive Disclaimer Appendix at the end of this document.

Finding Omics Data Assistant

The Omics Data Assistant is found in the Cohort Browser. The button is highlighted in navy in the figure below:

Using Omics Data Assistant to build a Cohort

1. Start by selecting the Omics Data Assistant button (highlighted above)

2. If this is the first time you are using the Omics Data Assistant, you will need to index that data. This step is done once. If the dataset needs to be index, you will have the following screen:

Select “start indexing” to index your data. You will have a notification on the screen with “Preparing the Dataset” and a progress bar.

1. Once your data has been indexed, you will be able to type in prompts. You will have a spot to write your own prompt to start the conversation, as well as sample prompts. At the bottom, there is your conversation history. The first screen will look like this:

1. Once you have finished the first prompt, the dialog box will appear like this:

You can utilize the tool to build cohorts and explore your dataset utilizing conversational language. When creating cohorts, you can also save the cohort and explore it within cohort browser, simply by following the prompts.

Comprehensive Disclaimer Appendix

1. Accuracy & Verification: ODA's responses are based on learned patterns and may be inaccurate or incomplete. All results must be independently verified by the user with domain expertise before use in scientific conclusions or clinical decisions. ODA is for preliminary exploration only.

2. Data Context & Scope: Complex prompts or terminology may yield unexpected results. ODA's output is limited strictly to indexed data. ODA is a research tool only and is not for clinical diagnosis or patient care. ODA uses conversations as context for the LLM but does not use them for model training or fine-tuning. Conversation history is stored within DNAnexus and accessible only to the user within the ODA interface.

3. User Responsibility: You are fully responsible for the proper use and interpretation of ODA's results. Developers are not liable for consequences from misuse or misinterpretation.

4. Non-Deterministic Results: Users should be aware that ODA may provide different responses to the same query and that LLM-generated summaries are probabilistic, not absolute.

5. Verification Requirement: All outputs, including statistical interpretations and gene-trait associations, must be verified by a qualified professional using primary data sources or raw code.

6. Limited Context: ODA analyzes data based on the provided parameters and cannot account for biological variables or metadata not explicitly uploaded or integrated.

7. This tool is for Research Use Only (RUO). ODA is not a medical device and has not been cleared by the FDA or any other regulatory body for clinical diagnostics or treatment decisions.

8. Not Medical Advice: ODA does not provide medical diagnoses, treatment recommendations, or prognostic assessments.

9. Experimental Nature: Insights generated (e.g., pathway enrichment or biomarker identification) are hypotheses for further experimental validation, not established biological facts.

10. Database Latency: ODA leverages a generic LLM that is trained using an internal knowledge base that has a "cutoff date" and may not reflect the most recent publications or updated genomic assemblies (e.g., GRCh38 vs. CHM13).

11. Users are strictly prohibited from uploading, entering, or otherwise providing Personally Identifiable Information (PII) or Protected Health Information (PHI) as defined by HIPAA, GDPR, or applicable local laws. This ODA is intended solely for the analysis of fully de-identified, pseudonymized, or synthetic research data.