Skill UI

Provides programmatic access to massive, versioned public single-cell and spatial transcriptomics datasets from CZ CELLxGENE. This skill allows users to efficiently query specific subsets of cell metadata, gene expression matrices, embeddings, and summary statistics across various organisms, tissues, and diseases. It is ideal for large-scale, memory-efficient biological research and model training without full dataset downloads.

Geniml is a comprehensive Python package designed for applying machine learning models to genomic interval data, typically stored in BED files. It provides several unsupervised methods—including Region2Vec and BEDspace—to learn robust embeddings of genomic regions, single cells (scATAC-seq), and associated metadata. Use cases include dimensionality reduction, structural variation analysis, cell-type annotation, and cross-modal querying in genomics research.

Molfeat is a comprehensive Python library designed for molecular featurization. It converts chemical structures (SMILES/RDKit) into numerical feature vectors, supporting over 100 featurizers like ECFP, MACCS, and deep embeddings (ChemBERTa). It unifies featurization pipelines, making it ideal for QSAR modeling, virtual screening, and integrating into deep learning ML workflows.

A comprehensive Python wrapper for OpenMS, designed for sophisticated computational mass spectrometry workflows. It supports end-to-end analysis in proteomics and metabolomics, enabling feature detection, peptide/protein identification, quantitative analysis (label-free/isobaric), and adduct/accurate mass annotation. Ideal for researchers processing complex LC-MS/MS raw data and running high-level pipelines.

PyTDC (Therapeutics Data Commons) is an open-science platform providing comprehensive, AI-ready datasets and benchmarks for the entire drug discovery and development pipeline. It supports single-instance prediction (e.g., ADME, toxicity), multi-instance prediction (DTI, DDI, PPI), and novel molecule generation (MolGen). Use this skill for benchmarking machine learning models and predicting complex biomedical properties.

SHAP (SHapley Additive exPlanations) provides a unified, robust framework for understanding complex machine learning models. It calculates feature importance using Shapley values from game theory, allowing users to interpret individual predictions, analyze model bias, visualize feature interactions, and debug model behavior. Applicable across deep learning, tree-based, and black-box models.

UMAP (Uniform Manifold Approximation and Projection) is a powerful technique for non-linear dimensionality reduction. This skill enables the creation of scalable, structure-preserving embeddings (2D/3D) for visualization, clustering preprocessing, and advanced machine learning feature engineering. It supports various metrics (e.g., Euclidean, Cosine) and supervised learning workflows.

This guide details how to configure and manage resource constraints in ClickHouse, addressing server-side limits, user quotas, and connection management. Use it when encountering "too many simultaneous queries," optimizing resource usage, or implementing robust rate limiting strategies in applications connecting to ClickHouse. Covers SQL quota setup and client-side pooling techniques.

A comprehensive skill designed for finance-related information retrieval. It provides dual search capabilities: real-time web searching across multiple engines (Jina, DDG, Baidu) for general market knowledge, and Retrieval-Augmented Generation (RAG) for querying local, proprietary document stores (e.g., daily news). Ideal for financial analysts needing both broad market context and specific internal data insights.

This skill manages the complexities of Clari API version changes, ensuring smooth transitions (e.g., v3 to v4). It guides developers through detecting schema drift, performing data field compatibility checks, and executing necessary database migrations (SQL/Python), ensuring reliable data pipelines after platform updates.

This guide details how to set up and execute large-scale, distributed GPU training jobs on the CoreWeave platform. It covers both single-node multi-GPU setups and multi-node training utilizing PyTorch DDP and Kubernetes resources (A100/H100). This workflow is ideal for fine-tuning large language models (LLMs) or running complex deep learning models that require high-performance, shared-storage compute clusters.

This skill provides comprehensive patterns for extracting and transforming Navan booking and transaction data. It addresses complex data pipeline challenges such as API pagination, date range chunking, and UUID-based deduplication. It supports both direct REST API implementation (via code examples) and configuration for managed connectors like Fivetran and Airbyte. Ideal for populating data warehouses, building robust analytics dashboards, or ensuring data quality from Navan sources.