Valérie Losier

A new paper submitted with ‏‏Adiel Ashrov‏ and ‏Guy Katz‏ which is the result of interesting research. How to measure and assess the resilience and robustness of AI models. This time, the focus is on LLM. #TRUST_AI #LLM #HUJI

4

Datavant's Nausheen Moulana discusses the early days of her engineering career, advice to others with similar goals, and how she was "immortalized in MATLAB code" Learn more: https://spr.ly/6042hC1rC

2

The article, "MISTRAL AI Agents for Protein Folding: A Conceptual Framework," outlines an AI system that utilizes a modular, multi-agent approach to address the complex challenge of protein folding. This approach employs several specialized AI agents, each equipped with distinct expertise and a set of tools. The modularity of the system offers significant advantages, including the division of labour, enhanced efficiency, and accuracy, as agents specialize in specific domains. This multi-agent paradigm reflects the collaborative nature of real-world scientific research. The system simulates a complex scientific workflow using a multi-agent system built with the Mistral AI SDK. The conceptual framework showcases specialized agents responsible for specific domain tasks: Protein Sequence Data Agent: Functions as a "biological librarian," swiftly fetching amino acid sequences and associated metadata for a given protein ID. This agent also identifies existing experimental 3D structures, such as those available in the Protein Data Bank (PDB). Folding Prediction & Simulation Agent: Conceptually simulates the dynamic process of folding. While a short amino acid sequence might be insufficient for a meaningful prediction, the agent can still outline the process of molecular dynamics simulation. Misfolding Analysis & Intervention Agent: Pinpoints "hotspots"—specific regions within a protein prone to misfolding or aggregation. Result Synthesis & Interpretation Agent: Weaves together predicted structures, folding dynamics, and misfolding analyses into a comprehensive report, transforming raw data and analytical insights into actionable knowledge. Historical & Ethical Context Agent: Provides a broader perspective by recalling significant milestones in protein science, such as Levinthal's paradox, and analyzes ethical implications, including concerns related to germline editing, accessibility, equity issues, and off-target effects. The agents are equipped with specific tools (implemented as mock functions in the demonstration) that allow them to perform domain-specific actions. The system manages conversation history and processes tool outputs to generate comprehensive responses, showcasing the orchestration and workflow of the MISTRAL AI system. The demonstration utilizes 'mock' functions to simulate the behaviour of complex scientific processes. #agenticai #mistral #opensource #generativeai #ai #proteomics

1