記事を投稿しました！ 🐾 【永久保存版】AITuberの「脳」を創る。致命的崩壊を防ぐ記憶システム・実装ロードマップ [rag] on #Qiita qiita.com/618kazeman/items/9…

📄🚨 Brand New Paper → Let’s Go From Paper to Practice: An impressive end-to-end Drug Discovery Multi-AI Agent that finds new medicines automatically. I show exactly how you can build it yourself. 👇 𝗪𝗵𝘆 𝘁𝗼 𝗕𝘂𝗶𝗹𝗱 𝗜𝘁 AI Agents like LIDDIA speed up discovery that normally takes months now done in hours. Tested on 30 human targets, it succeeded on 73%. finding 5 strong, diverse molecules per target, each passing drug-likeness rules and showing solid binding. Example: For a cancer-related protein (AR/NR3C4), it found a new molecule with excellent binding — all in silico (computer-based). Goal: build your own version that plans, generates, and improves molecules automatically. 𝗪𝗵𝗮𝘁 𝗬𝗼𝘂 𝗕𝘂𝗶𝗹𝗱 In drug discovery, data comes from many sources — proteins, molecules, and chemistry rules. You’ll build an AI Agent System that: ➡️ Understands the target ➡️ Creates molecule ideas ➡️ Tests and improves them ➡️ Repeats — until it finds winners Fast. Smart. Fully automated. 𝗦𝘁𝗲𝗽𝘀 1️⃣ Target Agent Sets what we’re looking for — the protein and what makes a good molecule (for example: strong binding, easy to make, drug-like). → Think: “Find 5 strong, diverse molecules for this protein.” → Build: Use a simple Planner Agent to define goals. 2️⃣ Pocket Agent Loads and prepares the protein pocket (from PDB files). → Think: cleaning and defining the active site. → Build: Small preprocessing script or 3D-prep agent. 3️⃣ Generator Agent Creates new molecules that might fit the pocket. → Think: hundreds of molecule ideas from AI models. → Build: Connect a generative model such as Pocket2Mol. 4️⃣ Evaluator Agent Checks each molecule’s quality. ✔ Drug-likeness (QED) ✔ Easy to make (SAS) ✔ Docking strength (Vina) ✔ Novelty and diversity → Think: automatic scoring with RDKit or Vina. → Build: an Evaluator Agent with simple metrics. 5️⃣ Filter Agent Keeps only the best and most unique molecules. → Think: “Keep top 10 diverse ones with good scores.” → Build: Filter logic with clustering and thresholds. 6️⃣ Optimizer Agent Improves promising molecules step by step — small changes to boost scores and lower binding energy. → Think: evolution of molecules guided by feedback. → Build: Optimization loop with GraphGA or similar. 7️⃣ Memory Planner Loop The system remembers what it tried before and plans the next move: Generate → Evaluate → Optimize → Repeat until it finds strong results or reaches its limit. → Think: a feedback loop with planning and memory. Paper: arxiv.org/pdf/2502.13959 ≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣≣ ⫸ꆛ Want to build Real-World AI Agents? Join My 𝗛𝗮𝗻𝗱𝘀-𝗼𝗻 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝟱-𝗶𝗻-𝟭 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 ! ➠ Build Agents for Healthcare, Finance, Smart Cities & More ➠ Master 5 Modules: 𝗠𝗖𝗣 · LangGraph · PydanticAI · CrewAI · OpenAI Swarm ➠ Includes 9 Full Projects 👉 𝗘𝗻𝗿𝗼𝗹𝗹 𝗡𝗢𝗪 (𝟱𝟲% 𝗢𝗙𝗙): maryammiradi.com/ai-agents-m…

This day 1988 - The Super Station launches with an overnight sustaining service from 10pm-6am. The jingles were produced by TM Productions and Muff Murfin. #AircheckDownloads