Remove Biomni due to incompatibilities

docs/scientific-skills.md

@@ -161,7 +161,6 @@
 - **Zarr** - Python library implementing the Zarr chunked, compressed N-dimensional array storage format. Provides efficient storage and access to large multi-dimensional arrays with chunking and compression. Key features include: support for NumPy-like arrays with chunked storage, multiple compression codecs (zlib, blosc, lz4, zstd), support for various data types, efficient partial array reading (only load needed chunks), support for both local filesystem and cloud storage (S3, GCS, Azure), and integration with NumPy, Dask, and Xarray. Enables working with arrays larger than available RAM through lazy loading and efficient chunk access. Supports parallel read/write operations and is optimized for cloud storage backends. Use cases: large-scale scientific data storage, cloud-based array storage, out-of-core array operations, and efficient storage of multi-dimensional datasets (genomics, imaging, climate data)
 
 ### Multi-omics & AI Agent Frameworks
-- **BIOMNI** - Autonomous biomedical AI agent framework from Stanford SNAP lab for executing complex research tasks across genomics, drug discovery, molecular biology, and clinical analysis. Combines LLM reasoning with code execution and ~11GB of integrated biomedical databases (Ensembl, NCBI Gene, UniProt, PDB, AlphaFold, ClinVar, OMIM, HPO, PubMed, KEGG, Reactome, GO). Supports multiple LLM providers (Claude, GPT-4, Gemini, Groq, Bedrock). Includes A1 agent class for autonomous task decomposition, BiomniEval1 benchmark framework, and MCP server integration. Use cases: CRISPR screening design, single-cell RNA-seq analysis, ADMET prediction, GWAS interpretation, rare disease diagnosis, protein structure analysis, literature synthesis, and multi-omics integration
 - **Denario** - Multiagent AI system for scientific research assistance that automates complete research workflows from data analysis through publication. Built on AG2 and LangGraph frameworks, orchestrates specialized agents for hypothesis generation, methodology development, computational analysis, and LaTeX paper writing. Supports multiple LLM providers (Google Vertex AI, OpenAI) with flexible pipeline stages allowing manual or automated inputs. Key features include: end-to-end research automation (data description → idea generation → methodology → results → paper), journal-specific formatting (APS and others), GUI interface via Streamlit, Docker deployment with LaTeX environment, reproducible research with version-controlled outputs, literature search integration, and integration with scientific Python stack (pandas, sklearn, scipy). Provides both programmatic Python API and web-based interface. Use cases: automated hypothesis generation from datasets, research methodology development, computational experiment execution with visualization, publication-ready manuscript generation, time-series analysis research, machine learning experiment automation, and accelerating the complete scientific research lifecycle from ideation to publication
 - **HypoGeniC** - Automated hypothesis generation and testing using large language models to accelerate scientific discovery. Provides three frameworks: HypoGeniC (data-driven hypothesis generation from observational data), HypoRefine (synergistic approach combining literature insights with empirical patterns through an agentic system), and Union methods (mechanistic combination of literature and data-driven hypotheses). Features iterative refinement that improves hypotheses by learning from challenging examples, Redis caching for API cost reduction, and customizable YAML-based prompt templates. Includes command-line tools for generation (hypogenic_generation) and testing (hypogenic_inference). Research applications have demonstrated 14.19% accuracy improvement in AI-content detection and 7.44% in deception detection. Use cases: deception detection in reviews, AI-generated content identification, mental stress detection, exploratory research without existing literature, hypothesis-driven analysis in novel domains, and systematic exploration of competing explanations