⭐️Mentor applications open⭐️ We're excited to announce that LOGML summer school will return in London: July 7-11 2025. We are seeking mentors to lead group projects at the intersection of geometry and machine learning. Find out more and apply:

(1/4) Excited to introduce ProCyon: a multimodal foundation model to model, generate, and predict protein phenotypes, led by stellar @oq_35 @YepHuang Robert Calef @valegiunca 👉 ProCyon is an 11B parameter multimodal model that integrates protein sequences, structures, and natural language ProCyon models, predicts, and generates phenotypes with zero-shot task transfer, breaking past the limits of fixed vocabularies or ontologies 🔬 The Challenge: Predicting protein structure is advanced, but predicting protein phenotypes—observable traits linking molecular function to biology—is still an open problem. 40% of human proteins lack context-specific insights, while 20% remain almost entirely uncharacterized

Introducing KGARevion, a KG+LLM agent 🤖designed for knowledge-intensive medical QA. Led by @xiaorui_su @GaoShanghua @valegiunca Here's why it is unique: 1️⃣ Open-ended reasoning: Generates nuanced answers without relying on predefined options 2️⃣ Robustness: LLMs can be surprisingly sensitive to how candidate answers are ordered and indexed in multi-choice setups Unlike LLMs, which show order sensitivity in multi-choice setups (favoring answers in the first position!), KGARevion is immune to this bias, ensuring more reliable selections 3️⃣ Flexibility: Combines LLM-generated triplets with KGs for verified answers. The multi-step process makes KGARevion adaptable to different reasoning styles, outperforming RAG models Medicine relies on various reasoning strategies, including rule-based, prototype-based, case-based, and analogy-based vertical reasoning (Patel et al., 2005). This variety calls for approaches that accommodate different styles and integrate specialized, in-domain knowledge. For instance, an organism such as Drosophila is used as an exemplar to model a disease mechanism, which is then applied by analogy to other organisms, including humans. In clinical practice, the patient serves as an exemplar, with generalizations drawn from many overlapping disease models and similar patient populations. 4️⃣ Versatility: Works with different LLMs and medical KGs. Preprint: Thanks Discover AI for highlighting KGARevion: @HarvardDBMI @harvard_data @KempnerInst

Excited to share our perspective in @CellCellPress, where we discuss “AI scientists” as collaborative AI agents designed to empower biomedical research While the concept of an “AI scientist” is aspirational, advances in agent-based AI are paving the way for AI agents as conversable systems with reflective and reasoning abilities that coordinate LLMs, ML tools, experimental platforms, or combinations thereof. We outline initial autonomy levels for agents based on proficiency in hypothesis generation, experimental design, execution, and reasoning: Level 0: No AI agents Level 1: AI agents as research assistants Level 2: AI agents as collaborators Level 3: AI agents as scientists These “AI scientists” could enhance discovery workflows by introducing skepticism and reasoning. Our vision is to amplify human creativity, enabling AI to handle large datasets, navigate complex hypotheses, and perform repetitive tasks faster. Many ethical considerations arise with biomedical AI agents. We discuss issues of governance, robustness, evaluation protocols, dataset generation, and associated risks. Imagine AI agents that help: 🔬 Design discovery workflows 🌱 Simulate virtual cells 🎛️ Control phenotypes programmatically ⚙️ Design cellular circuits 💊 Optimize therapeutic discovery and development Many thanks to fantastic group of co-authors @GaoShanghua, @AdaFang_ @YepHuang, @valegiunca, @ayushnoori, @schwarzjn_, @YEktefaie, @kondic_jovana @HarvardDBMI @harvardmed @KempnerInst @harvard_data @broadinstitute