Machine Learning Skills

Use this skill when the user wants to debug, diagnose, or systematically iterate on an experiment that already exists, or when they need a structured experiment log for tracking runs, hypotheses, failures, results, and next steps during active research. Apply it to underperforming methods, training that will not converge, regressions after a change, inconsistent results across datasets, aimless experimentation without progress, and questions like 'why doesn't this work?', 'no progress after many attempts', or 'how should I investigate this failure?'. Also use it for setting up practical experiment logging/record-keeping that supports debugging and iteration. Do not use it for designing a brand-new experiment pipeline or full experiment program (use experiment-pipeline), generating research ideas, fixing isolated coding/syntax errors, or writing retrospective summaries into research memory/notes/knowledge bases.

Infer gene regulatory networks (GRNs) from gene expression data using scalable algorithms (GRNBoost2, GENIE3). Use when analyzing transcriptomics data (bulk RNA-seq, single-cell RNA-seq) to identify transcription factor-target gene relationships and regulatory interactions. Supports distributed computation for large-scale datasets.

Molecular ML with diverse featurizers and pre-built datasets. Use for property prediction (ADMET, toxicity) with traditional ML or GNNs when you want extensive featurization options and MoleculeNet benchmarks. Best for quick experiments with pre-trained models, diverse molecular representations. For graph-first PyTorch workflows use torchdrug; for benchmark datasets use pytdc.

This skill should be used when working with genomic interval data (BED files) for machine learning tasks. Use for training region embeddings (Region2Vec, BEDspace), single-cell ATAC-seq analysis (scEmbed), building consensus peaks (universes), or any ML-based analysis of genomic regions. Applies to BED file collections, scATAC-seq data, chromatin accessibility datasets, and region-based genomic feature learning.

This skill should be used at the start of any computationally intensive scientific task to detect and report available system resources (CPU cores, GPUs, memory, disk space). It creates a JSON file with resource information and strategic recommendations that inform computational approach decisions such as whether to use parallel processing (joblib, multiprocessing), out-of-core computing (Dask, Zarr), GPU acceleration (PyTorch, JAX), or memory-efficient strategies. Use this skill before running analyses, training models, processing large datasets, or any task where resource constraints matter.

Lightweight WSI tile extraction and preprocessing. Use for basic slide processing tissue detection, tile extraction, stain normalization for H&E images. Best for simple pipelines, dataset preparation, quick tile-based analysis. For advanced spatial proteomics, multiplexed imaging, or deep learning pipelines use pathml.

Use when the user is doing AI/ML work in a scientific domain — biology, chemistry, physics, astronomy, climate, genomics, materials science, medicine, ecology, energy, conservation, engineering, mathematics, scientific reasoning, drug discovery, protein design, weather modeling, theorem proving, single-cell, PDE solving, or anything similar. Hugging Science (huggingscience.co) is a curated catalog of scientific datasets, models, blog posts, and interactive Spaces; the `hugging-science` org on Hugging Face hosts community datasets, models, and demo Spaces. This skill helps you discover the right resource AND actually use it — loading datasets via `datasets`, running models via `transformers` or the HF Inference API, calling Spaces like BoltzGen via `gradio_client`, and citing blog posts for methodology. Trigger this skill whenever a user mentions a scientific ML task, asks for "a dataset/model for X" where X is a scientific topic, wants to fine-tune on scientific data, asks about protein / molecule / genome / climate / materials / astronomy / pathology / weather ML, or needs AI tools for research — even if they never say "Hugging Science" explicitly. The catalog is purpose-built for LLM agents (it ships an `llms-full.txt`); prefer it over generic web search for these tasks.

Structured hypothesis formulation from observations. Use when you have experimental observations or data and need to formulate testable hypotheses with predictions, propose mechanisms, and design experiments to test them. Follows scientific method framework. For open-ended ideation use scientific-brainstorming; for automated LLM-driven hypothesis testing on datasets use hypogenic.

Query and download public cancer imaging data from NCI Imaging Data Commons using idc-index. Use for accessing large-scale radiology (CT, MR, PET) and pathology datasets for AI training or research. No authentication required. Query by metadata, visualize in browser, check licenses.

Neuropixels neural recording analysis. Load SpikeGLX/OpenEphys data, preprocess, motion correction, Kilosort4 spike sorting, quality metrics, Allen/IBL curation, AI-assisted visual analysis, for Neuropixels 1.0/2.0 extracellular electrophysiology. Use when working with neural recordings, spike sorting, extracellular electrophysiology, or when the user mentions Neuropixels, SpikeGLX, Open Ephys, Kilosort, quality metrics, or unit curation.

Microscopy data management platform. Access images via Python, retrieve datasets, analyze pixels, manage ROIs/annotations, batch processing, for high-content screening and microscopy workflows.

Full-featured computational pathology toolkit. Use for advanced WSI analysis including multiplexed immunofluorescence (CODEX, Vectra), nucleus segmentation, tissue graph construction, and ML model training on pathology data. Supports 160+ slide formats. For simple tile extraction from H&E slides, histolab may be simpler.

Hardware-agnostic quantum ML framework with automatic differentiation. Use when training quantum circuits via gradients, building hybrid quantum-classical models, or needing device portability across IBM/Google/Rigetti/IonQ. Best for variational algorithms (VQE, QAOA), quantum neural networks, and integration with PyTorch/JAX/TensorFlow. For hardware-specific optimizations use qiskit (IBM) or cirq (Google); for open quantum systems use qutip.

Fast in-memory DataFrame library for datasets that fit in RAM. Use when pandas is too slow but data still fits in memory. Lazy evaluation, parallel execution, Apache Arrow backend. Best for 1-100GB datasets, ETL pipelines, faster pandas replacement. For larger-than-RAM data use dask or vaex.

Build clinical/healthcare deep-learning pipelines with PyHealth — loading EHR/signal/imaging datasets (MIMIC-III/IV, eICU, OMOP, SleepEDF, ChestXray14, EHRShot), defining tasks (mortality, readmission, length-of-stay, drug recommendation, sleep staging, ICD coding, EEG events), instantiating models (Transformer, RETAIN, GAMENet, SafeDrug, MICRON, StageNet, AdaCare, CNN/RNN/MLP), training with the PyHealth Trainer, computing clinical metrics, and using medical code utilities (ICD/ATC/NDC/RxNorm lookup and cross-mapping). Use this skill whenever the user mentions PyHealth, MIMIC, eICU, OMOP, EHR modeling, clinical prediction, drug recommendation, sleep staging, medical code mapping, ICD/ATC codes, or any healthcare ML pipeline that fits the dataset → task → model → trainer → metrics pattern, even if "PyHealth" isn't named explicitly.

Therapeutics Data Commons. AI-ready drug discovery datasets (ADME, toxicity, DTI), benchmarks, scaffold splits, molecular oracles, for therapeutic ML and pharmacological prediction.

Create publication-quality scientific diagrams using Nano Banana 2 AI with smart iterative refinement. Uses Gemini 3.1 Pro Preview for quality review. Only regenerates if quality is below threshold for your document type. Specialized in neural network architectures, system diagrams, flowcharts, biological pathways, and complex scientific visualizations.

RNA velocity analysis with scVelo. Estimate cell state transitions from unspliced/spliced mRNA dynamics, infer trajectory directions, compute latent time, and identify driver genes in single-cell RNA-seq data. Complements Scanpy/scVI-tools for trajectory inference.

Model interpretability and explainability using SHAP (SHapley Additive exPlanations). Use this skill when explaining machine learning model predictions, computing feature importance, generating SHAP plots (waterfall, beeswarm, bar, scatter, force, heatmap), debugging models, analyzing model bias or fairness, comparing models, or implementing explainable AI. Works with tree-based models (XGBoost, LightGBM, Random Forest), deep learning (TensorFlow, PyTorch), linear models, and any black-box model.

Guide for building Graph Neural Networks with PyTorch Geometric (PyG). Use this skill whenever the user asks about graph neural networks, GNNs, node classification, link prediction, graph classification, message passing networks, heterogeneous graphs, neighbor sampling, or any task involving torch_geometric / PyG. Also trigger when you see imports from torch_geometric, or the user mentions graph convolutions (GCN, GAT, GraphSAGE, GIN), graph data structures, or working with relational/network data. Even if the user just says 'graph learning' or 'geometric deep learning', use this skill.

PyTorch-native graph neural networks for molecules and proteins. Use when building custom GNN architectures for drug discovery, protein modeling, or knowledge graph reasoning. Best for custom model development, protein property prediction, retrosynthesis. For pre-trained models and diverse featurizers use deepchem; for benchmark datasets use pytdc.

This skill should be used when working with pre-trained transformer models for natural language processing, computer vision, audio, or multimodal tasks. Use for text generation, classification, question answering, translation, summarization, image classification, object detection, speech recognition, and fine-tuning models on custom datasets.

Real-time hand detection in egocentric videos using victordibia/handtracking. Outputs bounding boxes for hands, specifically trained on EgoHands dataset. Supports video input/output with labeled hand boxes. Lightweight and fast for egocentric view applications.

Use this skill whenever the user wants an end-to-end workflow for the ABCD Study dataset, including download via NIMH Data Archive, BIDS organization, and multimodal processing of sMRI, fMRI, and dMRI. Triggers include: 'ABCD Study', 'ABCD data', 'process ABCD', 'ABCD fMRI', 'ABCD sMRI', 'ABCD diffusion', or any request to run the ABCD multimodal pipeline. This is the NeuroClaw dataset-orchestration layer for ABCD.

Use this skill whenever the user wants an end-to-end workflow for the ABIDE (Autism Brain Imaging Data Exchange) dataset, including download, BIDS organization, and processing of sMRI and rs-fMRI data. Triggers include: 'ABIDE', 'ABIDE data', 'process ABIDE', 'ABIDE fMRI', 'ABIDE sMRI', 'autism imaging', or any request to run the ABIDE pipeline. This is the NeuroClaw dataset-orchestration layer for ABIDE.

Use this skill whenever the user wants an end-to-end workflow for the ADHD-200 dataset, including download, BIDS organization, and processing of sMRI and rs-fMRI data. Triggers include: 'ADHD-200', 'ADHD200', 'process ADHD data', 'ADHD fMRI', or any request to run the ADHD-200 pipeline. This is the NeuroClaw dataset-orchestration layer for ADHD-200.

Use this skill whenever the user wants an end-to-end workflow for ADNI data (fMRI + T1), including BIDS preparation, fMRIPrep preprocessing, and DK68 ROI pipeline. This is the NeuroClaw dataset-orchestration layer for ADNI.

Use this skill whenever the user wants an end-to-end workflow for the AIBL (Australian Imaging, Biomarkers and Lifestyle) dataset, including data access guidance, BIDS organization, and multimodal processing of sMRI and PET (PiB, FDG, tau). Triggers include: 'AIBL', 'AIBL data', 'process AIBL', 'AIBL PET', 'AIBL MRI', or any request to run the AIBL multimodal pipeline. This is the NeuroClaw dataset-orchestration layer for AIBL.

Use this skill whenever the user wants an end-to-end workflow for the AOMIC (Amsterdam Open MRI Collection) dataset, including data access, BIDS organization, and multimodal processing of sMRI, rs-fMRI, and task-fMRI. Triggers include: 'AOMIC', 'AOMIC data', 'process AOMIC', 'AOMIC fMRI', 'AOMIC resting state', or any request to run the AOMIC multimodal pipeline. This is the NeuroClaw dataset-orchestration layer for AOMIC.

Use this skill whenever the user wants to automatically organize raw neuroimaging data (DICOM, NIfTI, EEG, etc.) into a valid BIDS (Brain Imaging Data Structure) dataset. Triggers include: 'organize to BIDS', 'BIDS organizer', 'convert to BIDS', 'BIDS conversion', 'bidsify', 'create BIDS dataset', 'raw data to BIDS', or any request to structure data according to BIDS specification.

Use this model doc whenever the user wants to run BrainNetworkTransformer for fMRI phenotype prediction, including data loading, training, and evaluation. BNT uses dense FC matrices (no PyG dependency) with DEC pooling + interpretable transformer encoder.

Use this model doc whenever the user wants to run BrainGNN for fMRI phenotype prediction, including graph construction, training, and evaluation. This document focuses on model-level usage and delegates upstream preprocessing to fmri-skill (and optionally hcpya-skill for HCP data).

Use this skill whenever the user wants an end-to-end workflow for the Cam-CAN (Cambridge Centre for Ageing and Neuroscience) dataset, including BIDS validation, multimodal processing of sMRI, rs-fMRI, task-fMRI, and MEG, phenotype extraction, and QC integration. Triggers include: 'Cam-CAN', 'CamCAN', 'process Cam-CAN data', 'Cam-CAN MEG', 'Cam-CAN fMRI', or any request to run the Cam-CAN multimodal pipeline.

Use this skill whenever the user wants to create, activate, list, export, update, clone, remove, or otherwise manage conda environments, or when a deep-learning / model skill requires a clean, isolated conda environment (e.g. 'create conda env for torch 2.3 cuda', 'export current env to yml', 'list all my conda envs', 'update packages in neuroclaw-dl', 'remove old env', 'clone env for reproducibility', 'install pytorch in new env'). Triggers include: 'conda create', 'conda env', 'make new environment', 'export yml', 'activate env', 'conda list envs', 'update conda env', 'clean environment', 'reproduce env', 'conda remove'. This skill is the mandatory gatekeeper for conda operations: it ALWAYS plans first, shows commands + risks + best practices, and waits for explicit user confirmation before executing anything.

Use this skill whenever the user wants an end-to-end workflow for the DMT-HAR-MED dataset (ds006644), including download, BIDS organization, and processing of rs-fMRI data from a psychedelic intervention study. Triggers include: 'DMT-HAR-MED', 'DMT HAR MED', 'ds006644', 'process DMT data', 'psychedelic fMRI', or any request to run the DMT-HAR-MED pipeline. This is the NeuroClaw dataset-orchestration layer for DMT-HAR-MED.

Use this skill whenever the user wants to perform standardized preprocessing of functional MRI (fMRI) and anatomical MRI data using fMRIPrep. Triggers include: 'fmriprep', 'fMRIPrep', 'fMRI preprocessing', 'BIDS fMRI', 'run fmriprep', 'preprocess bold', 'BOLD preprocessing', 'anatomical preprocessing', or any request involving BIDS-organized fMRI datasets.

Use this skill whenever the user wants to remove site/scanner/batch effects from neuroimaging features before running downstream models, run mega-analysis across multiple datasets, or evaluate models with leave-site-out / site-stratified protocols. Triggers include: 'harmonize', 'ComBat', 'CovBat', 'site effect', 'scanner effect', 'batch effect', 'leave-site-out', 'mega-analysis', 'multi-site', 'cross-site', 'neuroHarmonize'. This is a horizontal cross-cutting layer between dataset skills and model skills.

Use this skill whenever the user wants an end-to-end workflow for the Healthy Brain Network (HBN) dataset, including download, BIDS organization, and multimodal processing of sMRI, dMRI, rs-fMRI, task-fMRI, and EEG data. Triggers include: 'HBN', 'Healthy Brain Network', 'process HBN', 'HBN fMRI', 'HBN EEG', or any request to run the HBN multimodal pipeline. This is the NeuroClaw dataset-orchestration layer for HBN.

Use this skill whenever the user wants an end-to-end workflow for the HCP Aging (HCP-A) dataset, including dataset download, BIDS organization, and multimodal processing of sMRI, fMRI, and dMRI. Triggers include: 'HCP Aging', 'HCP-A', 'process HCP Aging data', 'HCP Aging sMRI fMRI', or any request to run the HCP-A multimodal pipeline.

Use this skill whenever the user wants an end-to-end workflow for the HCP Development (HCP-D) dataset, including dataset download, BIDS organization, and multimodal processing of sMRI, fMRI, and dMRI. Triggers include: 'HCP Development', 'HCP-D', 'process HCP Development data', 'HCP Development sMRI fMRI', or any request to run the HCP-D multimodal pipeline.

Use this skill whenever the user wants an end-to-end workflow for the HCP Early Psychosis (HCP-EP) dataset, including dataset download, BIDS organization, and multimodal processing of sMRI, fMRI, and dMRI. Triggers include: 'HCP Early Psychosis', 'HCP-EP', 'process HCP Early Psychosis data', 'HCP EP sMRI fMRI', or any request to run the HCP-EP multimodal pipeline.

Use this skill whenever the user wants an end-to-end workflow for the HCP Young Adult (HCP-YA / HCP1200) dataset, including dataset download, BIDS organization, and multimodal processing of sMRI, fMRI, and dMRI. Triggers include: 'HCP Young Adult', 'HCP-YA', 'HCP1200', 'process HCP data', 'HCP sMRI fMRI DTI', or any request to run the HCP-YA multimodal pipeline.

Use this model doc whenever the user wants to run IBGNN (Interpretable Brain Graph Neural Network) for fMRI phenotype prediction. IBGNN is a PyG-based GNN with a learnable MLP message function over [x_i, x_j, edge_attr], designed for connectome-based brain disorder analysis with post-hoc edge-mask explainer support.

Use this skill whenever the user wants an end-to-end workflow for the Motor Neuron Disease (MND) dataset from OpenNeuro ds005874, including BIDS validation, multimodal processing of rs-fMRI and task-fMRI, phenotype extraction, and QC integration. Triggers include: 'MND', 'Motor Neuron Disease', 'ALS', 'Amyotrophic Lateral Sclerosis', 'process MND data', 'MND fMRI', or any request to run the MND multimodal pipeline.

Use this skill whenever the user wants an end-to-end workflow for the Longitudinal MS Lesion Segmentation Challenge dataset, including data validation, multimodal processing of T1w, T2w, FLAIR, and PD, lesion segmentation, and QC integration. Triggers include: 'MS Lesion Challenge', 'MS Lesion', 'ISBI MS', 'longitudinal MS', 'multiple sclerosis lesion', or any request to run the MS lesion segmentation pipeline.

Use this skill whenever the user wants to run the NeuroSTORM multi-model fMRI platform: preprocessing, pretraining (MAE or contrastive), fine-tuning, inference, or benchmarking. It covers 8 built-in models — NeuroSTORM, SwiFT, BrainGNN, BrainNetworkTransformer (BNT), LG-GNN, Com-BrainTF, IBGNN, BrainNetCNN — across 3 input modalities (voxel 4D, ROI time series 2D, functional connectivity 2D). Triggers include: 'fMRI', 'NeuroSTORM', 'SwiFT', 'BrainGNN', 'BNT', 'BrainNetCNN', 'LG-GNN', 'Com-BrainTF', 'IBGNN', 'fMRI preprocessing', 'fMRI foundation model', 'ROI time series', 'functional connectivity', 'brain graph', 'HCP', 'ABCD', 'UKB', 'ADHD200', 'COBRE', 'UCLA', 'NSD', 'BOLD5000', 'disease diagnosis from fMRI', 'pretrain fMRI model', 'fine-tune fMRI', or any request involving .nii/.nii.gz fMRI volume files.

Use this skill whenever the user wants an end-to-end workflow for the Neuroimaging in Frontotemporal Dementia (NIFD) dataset, including BIDS validation, multimodal processing of sMRI, rs-fMRI, and dMRI, phenotype extraction, and QC integration. Triggers include: 'NIFD', 'frontotemporal dementia', 'FTD', 'bvFTD', 'PPA', 'process NIFD data', or any request to run the NIFD multimodal pipeline.

Use this skill whenever the user wants an end-to-end workflow for the Natural Scenes Dataset (NSD), including data access, BIDS validation, multimodal processing of task-fMRI and structural MRI, stimulus metadata extraction, and QC integration. Triggers include: 'NSD', 'Natural Scenes Dataset', 'process NSD data', 'NSD fMRI', 'visual neuroscience', or any request to run the NSD multimodal pipeline.