2024.07.23
AI agents, shared memory system, multi-purpose templates, and Insilico's curated database for enhanced academic writing.
Science42: DORA
Introducing the Draft Outline Research Assistant workflow:
In today's dynamic academic world, drafting scientific articles quickly is essential. On top of their daily tasks performing experimental, clinical, or bioinformatics research, scientists must also analyze the vast published literature and compose novel manuscripts. Insilico’s new product — Draft Outline Research Assistant, DORA — is built to help. DORA makes performing multiple research tasks easier: from disease hypothesis, target discovery, clinical trials analysis, and bioinformatics analysis, all the way to drafting scientific articles. Using predefined workflows and templates that allow multiple AI agents to spawn and perform domain-specific tasks, Science42: DORA automatically reads relevant articles, gathers knowledge on the subject, and summarizes the discovered evidence.
While DORA is a complex multi-model system that operates various research tools, its primary task is to help researchers and busy biomedical professionals draft a variety of documents and research papers. Researchers need tools that not only accelerate the writing process but also automate and streamline the search for scientific references and their incorporation into draft documents. DORA addresses these needs, enabling users to focus more on their research and less on time-intensive writing tasks.
DORA employs AI Agents to simultaneously perform multiple analytical tasks and generate scientific texts fully autonomously. These AI Agents use a suite of proprietary data-driven tools to perform various parts of the writing process, from drafting outlines and gathering data from published sources to analyzing the data with Insilico’s bioinformatics platforms.
I don’t think there is a better description of multi-agent research systems to date than the one proposed by Dr. Marinka Zitnik from Harvard in her paper Empowering Biomedical Discovery with AI Agents. DORA follows similar principles. While the agents follow the research workflow defined by the template with the objective of developing a research publication or another knowledge product and utilizing pre-defined research resources, databases, and domain-specific AI models, the system can be expanded to broader scientific exploration and genuine invention.
Alex Zhavoronkov, Ph.D.
Founder and CEO of Insilico Medicine
How Does DORA Work?
DORA is designed to simplify the drafting process, enabling users to prepare document drafts with just a few clicks. The journey begins with the selection of a template from our collection. Each template defines a set of steps DORA will take to prepare a draft, including which AI agents will be used in this particular experiment. They also include the structure of the final document, data sources, and writing guidelines.

The collaborative effort of multiple agents uncovers valuable insights that can serve as a foundation for broader scientific exploration. DORA is a catalyst for innovation, empowering researchers to push the boundaries of scientific discovery.
Template customization
After selecting the template, you can define the research topic, gene or disease you're investigating. You can even input your current ideas, data types, and other customized key points, and DORA’s Agents will smoothly incorporate them into the final draft.
To write a draft, DORA first creates a plan for each section and then writes the section itself. Specialized agents are tasked with drafting the outlines, ensuring they are comprehensive and suitable for the user’s defined inputs and initial ideas. Users can then further customize the plan as necessary. Adjusting the outline helps focus the section on the most important ideas.
DORA’s AI Agents composition
These AI agents use large language models (LLMs) built with our custom tools that retrieve relevant information from Insilico’s Data Warehouse. For each template, the set of agents is different. For example, a research article on Gene-Disease association evidence leverages domain-specific AI agents integrated with PandaOmics, our AI-driven platform for target discovery. These AI Agents contain tools to access an extensive database of processed omics datasets, curated knowledge graphs, and other validated internal sources useful for biomedical draft preparation.
While DORA's flexibility allows for the incorporation of users’ results, it can also autonomously guide research by independently gathering information from PandaOmics. PandaOmics AI Agents enable DORA to efficiently extract omics data: gene expression, methylation, and knowledge graph.
DORA also incorporates Insilico Medicine's cutting-edge Precious3GPT,
a multi-modal, multi-omics, multi-species language model for predicting aging-associated targets and therapeutic compounds from omics datasets. Precious3GPT agents guide DORA in discovering prospective molecular targets in a corresponding template.
Some AI Agents are powered by the Retrieval-Augmented Generation (RAG) from scientific publications databases. DORA’s AI Agents formulate multiple queries to the database, finding more relevant papers and inserting relevant citations into the draft. You can also ask DORA to find relevant papers for specific claims in your manuscript.
DORA’s pipeline
One of DORA's standout features is a shared memory system that ensures the coherence of all sections, enhancing the logical flow of the final document. We first create the sections that involve data collection and analysis. We then store relevant texts and conclusions in the AI Agents’ memory to draft descriptive sections, such as the Introduction and Abstract. While generating these sections, the AI Agents are instructed to refer to the shared memory to incorporate insights from the previous iterations. This pipeline guarantees that major results, insights, and conclusions are discussed across all interconnected sections.
After the generation is complete, you can make necessary adjustments to the text or introduce new references with the help of DORA’s AI agents.

DORA is designed to revolutionize the landscape of scientific research by streamlining the drafting process. DORA helps you concentrate on what truly matters — innovative thinking and groundbreaking discoveries.
12 Templates Available at launch
DORA can assist with a wide range of scientific texts, from brief reports such as conference abstracts or press releases to full-scale research papers.
DORA proposes the Scientific Essay and Review Paper templates for biologists, analysts, and academic researchers who need to collect and discuss key sources on a specific topic.

The Perspective Article template can benefit subject matter experts such as chemists drug hunters, and academic researchers. This template is perfect for those looking to express personal insights and arguments on specific topics. By providing their topic and hypothesis, users can craft well-argued articles with DORA's assistance.

Entrepreneurs, science communicators, and public relations specialists will find the Press Release template useful for creating concise and impactful press releases. By inputting the topic, abstract, journal's name, and organization, users can quickly generate press releases that highlight key research findings.

Biomedical researchers, health scientists, and chemist physicians investigating the effects of nutraceuticals and Traditional Chinese Medicine compounds can use the Nutraceutical & TCM Research Paper template. By providing their hypothesis, they can create comprehensive research articles that integrate new findings with existing knowledge.

Analysts, academic researchers, and conference presenters can use the Conference Abstract template to craft compelling abstracts for conference submissions. This template helps summarize pre-established results into concise and engaging abstracts.
The Сase report template will benefit clinical researchers, medical professionals, and biologists working on rare cases, unique diagnostic approaches, or personalized treatment management. By providing details about the disease, users can create detailed and informative case reports.

The Scientific Blog Post template is ideal for science bloggers, science communicators, and analysts looking to present research outcomes or early findings. By inputting the topic, users can craft succinct blog posts suitable for sharing with the scholarly community.

Investors, grant applicants, and research funding seekers can make use of the Grant Application template. This helps users create detailed grant applications that outline the project's objectives, methodology, and qualifications, enhancing their chances of securing financial support.

Academic researchers, scientists, and chemist drug hunters who explore their data and integrate it with existing knowledge will find the Research Article templates useful. With the help of DORA, one template allows you to create detailed and insightful research articles by providing a hypothesis. The other template is geared more towards biological research and enables you to delve deep into your research by specifying the gene and disease of interest.

Biomedical researchers, biologists, and pharmaceutical companies that focus on aging research can significantly benefit from the selection of the Prospective Geroprotectors Identification template. This template integrates the groundbreaking Precious3GPT model. Precious3GPT is a multi-modal, multi-omics, multi-species language model developed by Insilico Medicine, specifically designed to tackle aging-related tasks. By integrating Precious3GPT with DORA's AI Agents, the template enables the prediction of potential geroprotectors based on user-defined tissue and gender parameters. The result is a comprehensive research article that not only identifies prospective geroprotectors but also provides an in-depth analysis and discussion of the findings.
Papers Generated with DORA
In collaboration with Prof. Morten Scheibye-Knudsen, Center for Healthy Aging, Department of Cellular and Molecular Medicine, University of Copenhagen, DORA was used to draft a primary research article centered on understanding the molecular signature underlying the variability in radiotherapy outcomes across cancer types with a focus on Glioblastoma Multiforme and Low-Grade Gliomas. This article is deposited in MedRxiv and can be accessed here: Read paper →
"DORA played a pivotal role in the latest research collaboration with Atossa Therapeutics, and Insilico Medicine. Atossa Therapeutics, under the visionary leadership of its founder Dr. Steven Quay, has been at the forefront of developing innovative treatments for breast cancer. Dr. Quay, a renowned expert in the field with a remarkable track record of pioneering research and numerous patents guided the team in leveraging DORA’s advanced scientific text generation and curated the article. Remmel, L., et al., ‘Comparative Analysis of Endoxifen, Tamoxifen, and Fulvestrant: A Bioinformatics Approach to Uncover Mechanisms of Action in Breast Cancer.’ The scope of the primary research article leveraged DORA's understanding of advanced bioinformatic approaches to elucidate the distinct mechanisms of action and potential therapeutic benefits of anti-estrogens therapies, offering insights into potential personalized treatment strategies. The article is set to be submitted for peer review in top-tier oncology journals.": Read paper →
Coming Soon

In collaboration with Dr. Filippo Castiglione, Executive Director, Biotechnology at the Technology Innovation Institute, Abu Dhabi, DORA was used to draft a primary research article that will be submitted to bioRxiv. Singh A., Castiglione F, et al., ‘Designing a multi-serotype Dengue virus vaccine: an in-silico approach to broad-spectrum immunity’. This work describes the development of a bioinformatics pipeline utilizing immune-related machine learning tools to identify epitopes and peptides of the Dengue virus that are useful to construct a multi-epitope vaccine to target multiple serotypes of the virus.
Coming Soon

Dr. Alessandro Luciani, group leader for Mechanisms of inherited kidney disorders, Institute of Physiology in the University of Zurich, used DORA to draft a comprehensive review focused on the FDA-approved compounds that may be repurposed for use in the treatment of cystinosis. This paper is planning to be submitted for peer review at a leading nephrology journal, Kidney International.
Coming Soon

DORA was central to drafting the latest research output from the established collaborations between Prof. Evgeny Izumchenko, Assistant Professor of Medicine, Committee on Genetics, Genomics and Systems Biology at the University of Chicago and Insilico Medicine. Prof. Izumchenko is an internationally recognized cancer biologist who authored multiple high-impact articles focused on head and neck cancer using state-of-the-art AI-driven computational approaches. The scope of the primary research article drafted by DORA will uncover novel targets combined with personalized medicine approaches. It will be submitted for peer review at leading oncology-oriented journals.
Roadmap
for DORA
The development of DORA doesn’t stop here. In the upcoming weeks, we will be releasing multiple important updates based on user feedback to make DORA even more useful and responsive to your needs. Your feedback and suggestions are crucial for shaping the next steps for DORA.
1
Customize bibliography
for document generation
2
Enhance exported document
by adding custom bibliography details
3
Improve document's coherence with AI
after manual document editing
4
Enhance document customization
by including custom sections
5
Enrich data sources
for more areas and domains
6
Adhere to journals' content guidelines
with journal-specific templates
Stay tuned, follow us on social media!