データ・AIのスキル

Prepares statistical reports for clinical trials following CONSORT 2025, SPIRIT 2025, ICH E9(R1) estimands, and FDA 2023 covariate adjustment guidance. Covers Table 1 generation, analysis populations (ITT/FAS/PP/Safety), the 5 ICH E9(R1) intercurrent-event strategies, MMRM under MAR (mmrm), reference-based MI (rbmi J2R/CR/CIR), Permutt tipping-point sensitivity, and Rubin's-rules vs frequentist variance debate. Use when preparing regulatory submissions, defining estimands, or implementing missing-data sensitivity analyses.

日本語の概要は準備中です。原文の説明を表示しています。

Performs logistic regression for clinical trial outcomes (binary, ordinal, multinomial) with marginal-vs-conditional estimand reporting per FDA 2023 covariate adjustment guidance, g-computation/standardisation for marginal effects, modified Poisson for RR, Brant test for proportional odds, Firth penalty for separation, and Hauck-Donner detection. Use when modeling binary or ordinal endpoints in confirmatory or exploratory clinical trials.

日本語の概要は準備中です。原文の説明を表示しています。

Implements multiplicity control for confirmatory clinical trials using graphical procedures (Bretz-Maurer-Hommel), gatekeeping (parallel, serial, mixed), Hochberg/Hommel/Holm with PRDS, and the closed-testing principle (Marcus-Peritz-Gabriel; Goeman 2021 admissibility). Covers FDA Multiple Endpoints Final Guidance (October 2022), graphical procedures via R gMCP, primary + key-secondary + subgroup hierarchies, and FWER vs FDR distinction. Use when designing the multiplicity strategy for confirmatory trials with multiple primary or key secondary endpoints.

日本語の概要は準備中です。原文の説明を表示しています。

Computes sample size and power for clinical trials including continuous, binary, and time-to-event endpoints; superiority, non-inferiority, and equivalence designs; FDA 2016 non-inferiority margin selection with M1/M2 framework; Schoenfeld 1981 and Lakatos 1988 for survival; Schuirmann TOST and 80-125% bioequivalence; minimum clinically important difference (MCID) vs δ distinction. Use when justifying trial size in protocol or SAP per CONSORT 2025 item 7.

日本語の概要は準備中です。原文の説明を表示しています。

Performs subgroup and heterogeneous treatment effect (HTE) analyses for clinical trials. Covers Mantel-Haenszel pooling, Breslow-Day, interaction tests in regression, RERI for additive interaction, modern data-adaptive HTE methods (STEPP, SIDES, causal forests, X/R-learners), Bayesian shrinkage (Dixon-Simon, MAP, EXNEX), graphical multiplicity (Bretz-Maurer), and credibility frameworks (Sun BMJ, EMA 2019). Use when analyzing treatment effects across patient subgroups for regulatory submissions or precision-medicine claims.

日本語の概要は準備中です。原文の説明を表示しています。

Performs time-to-event analysis for clinical trials including Cox proportional hazards regression with PH diagnostics, restricted mean survival time (RMST) under non-PH, competing risks via Fine-Gray vs cause-specific Cox, weighted log-rank and MaxCombo for non-proportional hazards, recurrent events (Andersen-Gill, PWP, WLW), and interval-censored data. Use when analyzing time-to-event endpoints (OS, PFS, DOR, TTR, TTNT) in oncology or other clinical trials.

日本語の概要は準備中です。原文の説明を表示しています。

Statistical modeling — linear/logistic/ordinal/Poisson regression, ANOVA, Kruskal-Wallis, chi-square, Mann-Whitney, Cox survival, spline fits (R `ns()`), odds ratios, Cohen's d, F-statistic, p-value computation. Specializes in clinical-trial AE analysis (SDTM DM/AE), severity ordinal regression, and per-feature stat workflows.

日本語の概要は準備中です。原文の説明を表示しています。

Plot per-group distributions of continuous data using boxplots, violins, beeswarms, quasirandom jitter, and raincloud plots with sample-size honesty (Weissgerber 2015), KDE-bandwidth awareness, and N-aware encoding choices. Use when comparing distributions across a small number of groups — expression per cluster, biomarker per arm, scores per condition — and the bar-of-mean default is misleading.

日本語の概要は準備中です。原文の説明を表示しています。

Build forest plots (HR, OR, RR, beta-coefficient summaries with CIs) and funnel plots (meta-analysis publication-bias diagnostics) using forestplot, metafor, ggforest, and MendelianRandomization with proper axis-scaling, summary-diamond placement, subgroup nesting, and Egger / trim-and-fill asymmetry tests. Use when summarizing effects across subgroups, trials, or instruments — meta-analysis, Mendelian randomization, subgroup HRs.

日本語の概要は準備中です。原文の説明を表示しています。

Build genome-browser-style multi-track figures with pyGenomeTracks (config-driven), Gviz (R), and IGV batch screenshotting. Covers BigWig coverage tracks, BED/peak overlays, gene-model rendering, Hi-C matrix tracks, BedPE link arcs, spike-in-aware normalization, and the bamCoverage --normalizeUsing trap. Use when producing publication figures of genomic loci with stacked aligned tracks (coverage, peaks, genes, interactions) for ChIP-seq, ATAC-seq, RNA-seq, Hi-C, or generic locus visualization.

日本語の概要は準備中です。原文の説明を表示しています。

Build publication-quality figures in R with ggplot2 using the grammar of graphics (data + aesthetics + geometries + scales + facets + themes) with CVD-safe palettes, cairo_pdf TrueType embedding, programmatic aes via tidy evaluation, and the theme_classic publication baseline. Use when producing static figures in R for papers, presentations, or reports.

日本語の概要は準備中です。原文の説明を表示しています。

Build clustered heatmaps for expression matrices and other features-by-samples data with rigorous distance/linkage/scaling choices, robust color mapping, optimal leaf ordering, and ComplexHeatmap/pheatmap/seaborn rendering. Covers the ward.D vs ward.D2 trap, the row-vs-column scaling decision, multi-track annotations, oncoPrint, and raster rendering for large matrices. Use when visualizing expression patterns across samples or identifying co-regulated clusters.

日本語の概要は準備中です。原文の説明を表示しています。

Build Sankey, alluvial, river, and CONSORT-style flow diagrams to visualize cohort transitions, cell-state changes, or pipeline filtering using ggalluvial, networkD3, plotly, and consort. Use when showing how entities move between categories across timepoints (cell states, drug response classes, patient flow through a trial) or filtering pipelines (variants filtered through QC stages).

日本語の概要は準備中です。原文の説明を表示しています。

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.

日本語の概要は準備中です。原文の説明を表示しています。

Rigorous methodology for evaluating ML models on established benchmarks. Covers proper train/val/test splits, baseline verification from original papers, exact metric formula discrepancies, data-leak detection checklist, multi-seed robustness, and honest reporting templates. Use when claiming to beat published baselines, writing methods papers, or auditing existing results.

日本語の概要は準備中です。原文の説明を表示しています。

GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristics), semantic deduplication, PII redaction, NSFW detection. Scales across GPUs with RAPIDS. Use for preparing high-quality training datasets, cleaning web data, or deduplicating large corpora.

日本語の概要は準備中です。原文の説明を表示しています。

High-performance reinforcement learning framework optimized for speed and scale. Use when you need fast parallel training, vectorized environments, multi-agent systems, or integration with game environments (Atari, Procgen, NetHack). Achieves 2-10x speedups over standard implementations. For quick prototyping or standard algorithm implementations with extensive documentation, use stable-baselines3 instead.

日本語の概要は準備中です。原文の説明を表示しています。

Expert guidance for Fully Sharded Data Parallel training with PyTorch FSDP - parameter sharding, mixed precision, CPU offloading, FSDP2

日本語の概要は準備中です。原文の説明を表示しています。

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

日本語の概要は準備中です。原文の説明を表示しています。

Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.

日本語の概要は準備中です。原文の説明を表示しています。

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.

日本語の概要は準備中です。原文の説明を表示しています。

Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform

日本語の概要は準備中です。原文の説明を表示しています。

Cost modeling and ROI analysis for specialized LLM development. Use when deciding whether to train a custom model, estimating total cost, or calculating break-even vs frontier APIs. Covers training costs, inference costs, and time-to-ROI projections.

日本語の概要は準備中です。原文の説明を表示しています。

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

日本語の概要は準備中です。原文の説明を表示しています。