Draft:Conversational genomics

Submission declined on 20 July 2026 by ChrysGalley (talk). This draft appears to contain text generated by a large language model (such as ChatGPT). You cannot use LLMs to generate article content. LLM-generated pages with certain obvious signs of being machine generated may be deleted without notice. Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject. See the advice page on large language models for more information. If you would like to continue working on the submission, click on the "Edit" tab at the top of the window. If you have not resolved the issues listed above, your draft will be declined again and potentially deleted. If you need extra help, please ask us a question at the AfC Help Desk or get live help from experienced editors. Please do not remove reviewer comments or this notice until the submission is accepted. Where to get help If you need help editing or submitting your draft, please ask us a question at the AfC Help Desk or get live help from experienced editors. These venues are only for help with editing and the submission process, not to get reviews. If you need feedback on your draft, or if the review is taking a lot of time, you can try asking for help on the talk page of a relevant WikiProject. Some WikiProjects are more active than others so a speedy reply is not guaranteed. How to improve a draft Wikipedia:Contributing to Wikipedia – a basic overview on how to edit Wikipedia. Help:Wikitext – how to use the markup Help:Referencing for beginners – how to include references Wikipedia:Article development – how to develop your article Wikipedia:Writing better articles – how to improve your article Wikipedia:Verifiability – make sure your article includes reliable third-party sources You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Add tags to your draft Editor resources Easy tools: Citation bot (help) | Advanced: Fix bare URLs Declined by ChrysGalley 30 days ago. Last edited by JCW-CleanerBot 5 days ago. Resubmit Please note that if the issues are not fixed, the draft will be declined again.

Submission declined on 20 July 2026 by ChrysGalley (talk). This draft appears to contain text generated by a large language model (such as ChatGPT). You cannot use LLMs to generate article content. LLM-generated pages with certain obvious signs of being machine generated may be deleted without notice. Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject. See the advice page on large language models for more information. Declined by ChrysGalley 30 days ago.

• Comment: Declining a second time since it didn't go to the submitter ChrysGalley (talk) 20:43, 20 July 2026 (UTC)

• Comment: This gets to be a circular logic at some point, but still, it's rather obviously AI generated. ChrysGalley (talk) 20:40, 20 July 2026 (UTC)

This article has multiple issues. Please help improve it or discuss these issues on the talk page. (Learn how and when to remove these messages) This article may incorporate text from a large language model, which is prohibited in Wikipedia articles. It may include hallucinated information, copyright violations, claims not verified in cited sources, original research, or fictitious references. Any such material should be removed. (July 2026) (Learn how and when to remove this message) The topic of this draft may not meet Wikipedia's notability guideline for neologisms. Please help to demonstrate the notability of the topic by citing reliable secondary sources that are independent of the topic and provide significant coverage of it beyond a mere trivial mention. Notability should be established before publishing this draft or submitting it via the articles for creation process. Find sources: "Conversational genomics" – news · newspapers · books · scholar · JSTOR (July 2026) (Learn how and when to remove this message) This article may contain original research. Please improve it by verifying the claims made and adding inline citations. Statements consisting only of original research should be removed. (July 2026) (Learn how and when to remove this message) (Learn how and when to remove this message)

Conversational genomics is a descriptive label for the use of natural-language conversational artificial intelligence — including chatbots and large language model (LLM) based agents — to query, interpret, and communicate genomic and clinical genetics information. Rather than requiring users to write code or navigate specialised software, such systems let clinicians, researchers, and patients pose questions in ordinary language and receive summaries, risk assessments, or the results of underlying analyses.

The exact phrase "conversational genomics" is not yet an established term in the peer-reviewed literature; it appears prominently in an industry blog post,^[1] while closely related work is published under other names such as "conversational agent",^[2] "conversational voice interface",^[3] and "generative AI in genetic counseling".^[4] The best-documented applications fall into two broad areas: genetic counseling and patient communication, where conversational agents have been studied in randomized controlled trials; and variant and sequence interpretation, where generative-AI assistants summarise evidence or wrap machine-learning models of the genome.

Interest in conversational interfaces for genomics has been driven partly by a shortage of genetic counselors relative to demand, and by the volume and complexity of genomic data that clinicians and patients must interpret.^[5][4] Two technical developments underpin the field.

The first is the maturation of general-purpose large language models capable of holding a dialogue and answering domain questions. Independent evaluations have tested such models on genetics tasks: a comparison of ChatGPT 3.5 and ChatGPT 4 on 68 patient-facing genetics questions found GPT-4 more accurate (mean 4.17 versus 3.38 out of 5) but still prone to outdated information and errors in inheritance calculations.^[5] Another benchmark evaluated multiple models on identifying 63 genetic conditions from clinician versus layperson descriptions, finding that closed-source models performed best (GPT-4 around 89–90%) but that accuracy dropped on real-world patient descriptions.^[6]

The second is a distinct line of work on sequence-to-function models — deep neural networks that predict molecular activity directly from DNA sequence. Examples include Enformer, which predicts gene expression and regulatory signals from roughly 200 kilobases of sequence context and scores non-coding variant effects;^[7] Borzoi, which predicts cell-type-specific RNA-seq coverage;^[8] and AlphaGenome, a unified model operating at single-base resolution over a 1-megabase context.^[9] These models are not themselves conversational, but conversational-genomics systems increasingly use them as back-end tools.

Reviews distinguish these sequence-based "genome language models", which treat nucleotides as tokens, from natural-language chat and agent systems.^[10][11] A related development is the emergence of "agentic" AI systems that combine an LLM with planning, memory, and the ability to invoke external tools. Surveys catalogue dozens of such systems across genomics and biomedicine and enumerate recurring challenges, including unstable reasoning, limited biological grounding, reproducibility, and safety.^[12][13] Proposals such as adapting the Model Context Protocol to bioinformatics web services aim to let LLM agents query databases such as GEO, STRING, and the UCSC Cell Browser autonomously.^[14]

An early peer-reviewed example of conversational access to genomic data is Melvin, a voice interface built as an Amazon Alexa skill that answers spoken questions about cancer genomics data from The Cancer Genome Atlas.^[3] More recent research prototypes couple an LLM with variant-annotation resources; a Google Cloud engineering blog described a multi-agent prototype using Gemini together with the Variant Effect Predictor and gnomAD data to answer natural-language variant-interpretation questions.^[1]

The most developed and rigorously studied application of conversational genomics is in genetic counseling, particularly for hereditary breast and ovarian cancer (HBOC) and other hereditary-cancer syndromes. A 2023 systematic review and meta-analysis in JCO Clinical Cancer Informatics identified seven studies evaluating five distinct chatbots for genetic cancer risk assessment and counseling, reporting high user acceptability and engagement, a pooled completion rate for risk assessment of about 37%, and no association between sociodemographic factors and interaction patterns; it also noted that comparative-effectiveness evidence was limited.^[4]

The largest randomized evidence comes from the BRIDGE trial, an equivalence-design randomized controlled trial of 3,073 patients at two US health systems. It found that a chatbot delivering pretest cancer-genetics education achieved completion of pretest genetic services and genetic testing equivalent to appointments with certified genetic counselors (difference 2.0 percentage points, 95% confidence interval −1.1 to 5.0).^[15] Earlier feasibility work described real patient interactions with the same automated conversational agent delivering pretest education.^[16] A smaller RCT of 37 women newly diagnosed with breast cancer found patient satisfaction and comprehension with chatbot pretest counseling comparable to in-person counseling.^[17]

Conversational agents have also been used for family-history collection and risk triage before clinical visits. A real-world deployment of the Gia chatbot recorded 61,070 users completing hereditary-cancer risk assessments before routine women's-health visits, with 89.4% completing the assessment and about 27% meeting testing criteria; the study's authors were affiliated with a commercial genetics-testing company.^[18] A separate clinical study of an AI chatbot performing preliminary HBOC screening in a small sample (n=11) reported that its determinations of whether users met NCCN BRCA1/2 testing criteria were consistent with certified genetic counselors' assessments.^[19] A qualitative implementation study examined barriers and facilitators to adopting such a screening chatbot in women's-health clinics.^[20]

Surveys of practising genetic counselors indicate that chatbot use to date has centred on communication with at-risk family members and patient education.^[4] Reviews and perspectives caution that most evidence remains at the feasibility and acceptability stage, with only one large equivalence RCT reported so far.^[4][15]

A second application area is assisting the interpretation and annotation of genomic variants. VarChat is a generative-AI assistant that takes a variant identifier (in HGVS nomenclature or as a dbSNP identifier) and returns a human-readable summary of the relevant literature with linked references, using retrieval-augmented generation and integrating ClinVar data.^[21] Other tools address related tasks: AI-CURA is an LLM workflow reported to match clinical experts on ACMG/AMP variant classification for a set of 150 variants,^[22] and AutoPM3 extracts a specific ACMG evidence code from the literature.^[23] Evaluations of general-purpose models report a mixed picture: one study found ChatGPT accurate on general genetic-testing counseling questions but less accurate on syndrome-specific ones (88.2% correct or comprehensive for HBOC versus 66.6% for Lynch syndrome).^[24]

Systems that couple a conversational interface directly to predictive models of the genome have also been described. ChatNT is a multimodal conversational agent that formulates supervised genomics tasks across DNA, RNA, and protein as text-to-text problems, combining a DNA sequence encoder with an LLM decoder so that users can query biological sequences in English.^[2] Chorus is an open-source system that exposes multiple sequence-to-function models — including Enformer, Borzoi, and AlphaGenome — through a natural-language and Model Context Protocol interface for variant-effect prediction.^{[note 1]} Independent surveys frame such tools within a broader landscape of generative-AI applications in medical genomics spanning variant identification, annotation, interpretation, and report generation.^[25][26]

Independent reviews and evaluations identify several recurring limitations. General-purpose conversational models can produce outdated or incorrect answers, with accuracy declining on syndrome-specific detail and on descriptions phrased by laypeople rather than clinicians.^[5][6][24] Agentic systems face challenges of unstable reasoning, hallucination, limited biological grounding, reproducibility, and biosafety.^[12][13] In the clinical-genetics setting, the evidence base is dominated by feasibility and acceptability studies, and some large real-world deployments were reported by teams with commercial interests in genetic testing.^[4][18] Implementation studies note practical barriers to adoption in routine care.^[20]

• DNA large language model
• Large language model
• Genetic counseling
• Model Context Protocol

Category:Artificial intelligence Category:Genomics Category:Bioinformatics Category:Genetic counseling Category:Chatbots Category:Large language models