Project Themes:

• Virtual AI Co-scientist

Team Lead(s):

• Name: Arvind Ramanathan
• Affiliation: Argonne National Laboratory
• Email: ramanathana@anl.gov

Suggested Team Members and Roles [4-6 members]

Name	Affiliation	Role / Expertise
Arvind Ramanathan	Argonne National Laboratory	Principal Investigator, AI Systems Architect
Archit Vasan	Argonne National Laboratory	Model integration and deployment
Bharat Kale	Argonne National Laboratory	Use-case definition and validation
Brian Hsu	Argonne National Laboratory	Model integration and deployment
Ozan Gokdemir	Argonne National Laboratory	RAG and Data Integration Specialist
Ruijie Zhu	University of Chicago	Use-case definition and validation

Project Summary

This project extends the open-source Co-Scientist framework (https://github.com/acadev/Jnana) to build reasoning agents for protein function prediction, using BV-BRC datasets as testbeds. The goal is to enable hypothesis-driven dialogue and AI-assisted exploration of uncharacterized genes and proteins relevant to priority pathogens.

Goals and Objectives

• Goal 1: Deploy Co-Scientist framework in the BV-BRC environment
• Goal 2: Implement a reasoning module for protein function prediction using multi-modal evidence (sequence, structure, literature)
• Goal 3: Demonstrate AI-guided hypothesis generation for uncharacterized genes in one pathogen model

Approach

The Co-Scientist framework will use an agentic reasoning loop combining retrieval (via HiPerRAG), synthesis (via Rhea), and hypothesis refinement through dialogue. The system will be trained on BV-BRC pathogen data and evaluated for its interpretability and accuracy in protein function annotation.

Data and Resources Required

Resource Type	Source / Link	Description / Purpose
Data	BV-BRC genomes and curated datasets	Training and evaluation data
Tools / Services	Jnana (Co-Scientist GitHub repo)	Agentic reasoning system
LLMs / AI Models	GPT-4, Claude 3	Reasoning and synthesis agents
Compute / Storage	Argonne HPC	Experimentation environment

Expected Outcomes / Deliverables

Prototype AI Co-Scientist agent capable of hypothesis-driven protein function prediction and reasoning trace visualization.

Potential Impact and Next Steps

This project showcases autonomous scientific reasoning and interactive discovery for infectious disease research. It will help BV-BRC develop transparent, AI-assisted research workflows and foster a foundation for broader agentic science platforms.

Technical Support Needed

• GPU / LLM access
• Mentor support
• API keys

Additional Comments