Claude Skills by HolobiomicsLab

Use when you have performed lazy dask-backed feature extraction on an ImageContainer using im.calculate_image_features and need to persist the computed spatial features into the AnnData object while controlling when and how dask graph computation is triggered.

Use when you have transcript-level abundance, count, and length estimates (from salmon, Sailfish, or kallisto via tximport) and want to perform gene-level differential expression analysis in DESeq2 using the 'original counts and offset' method, which accounts for changes in effective gene length.

Use when you have transcript-level quantification files (quant.sf, kallisto abundance.h5, or RSEM output) from one or more RNA-seq samples and need to construct a count matrix for differential expression analysis.

Use when when you have microarray or RNA-seq expression data paired with phenotype/sample metadata describing experimental conditions, treatments, or group assignments, and you need to fit a linear model to test for differential expression across those conditions.

Use when after fitting a linear model with lmFit on voom-transformed or log2-normalized RNA-seq or microarray expression matrices, apply eBayes moderation to moderate gene-wise variance estimates before extracting top differentially expressed genes.

Use when you have fitted a linear model to gene expression data (microarray, RNA-seq, qPCR, or proteomics) across multiple samples and need to compute gene-level test statistics.

Use when when you have RNA-seq count matrices (from HTSeq, featureCounts, Salmon, kallisto, or RSEM quantification) and need to test for differential expression between two or more treatment groups.

Use when you have PSI (percent-spliced-in) matrices for multiple samples grouped into two or more biological conditions, along with corresponding transcript expression quantification, and need to identify events or transcripts with statistically significant changes in inclusion levels between.

Use when apply PCA when you have a log-normalized, scaled gene expression matrix from highly variable genes and need to reduce dimensionality before constructing k-nearest neighbor graphs or other manifold-learning steps.

Use when you have a normalized single-cell expression matrix (e.g., after SCTransform) and need to compute gene-level covariance structure for pathway enrichment analysis (e.g., GESECA) rather than cell-level dimensionality reduction.

Use when you have raw RNA-seq count data (from HTSeq, featureCounts, Salmon, or similar quantification tools) organized in a count matrix with samples as columns and genes as rows, and you need to test for differential expression between conditions using a negative binomial model.

Use when you have transcript-level abundance estimates and count matrices from tximport (derived from Salmon, Sailfish, or kallisto output) and need to prepare them for differential expression analysis in edgeR.

Use when you have a fitted linear model (lmFit object) from microarray or RNA-seq count data and need to compute stable variance estimates and differential expression statistics despite having few biological replicates or small numbers of arrays.

Use when when constructing k-nearest neighbor graphs from spatial coordinates (e.g., microscopy x,y positions or tissue section coordinates) and exact neighbor discovery is computationally prohibitive;

Use when when evaluating a scientific manuscript that describes a computational analysis (e.g., a normalization or transformation pipeline), use this skill to trace each claimed finding to its source code, data, or prior publication.

Use when when a paper describes a computational or statistical method and you need to verify that claims are supported by available code, data, or documentation before human expert evaluation.

Use when when setting up a DESeqDataSet from count matrices or transcript quantification, you must specify a design formula before running DESeq() if your experiment has batch effects, multiple treatment groups, or multi-factor designs.

Use when after running DESeq() on a DESeqDataSet and extracting results with results(dds), you have a results table with log₂ fold changes, p-values, and adjusted p-values (padj). Use this skill when you need to (1) identify genes meeting a significance threshold (e.g., padj < 0.

Use when you have normalized or voom-transformed gene expression counts/intensities indexed by gene and sample, along with an experimental design matrix specifying condition, batch, or covariate assignments.

Use when after normalizing total UMI counts per cell using normalize_total, and before PCA or feature selection. Use this when working with raw or depth-normalized count matrices where gene expression values span multiple orders of magnitude and variance is not homogeneous across expression levels.

Use when when you have raw or unnormalized gene expression data from microarray experiments (e.

Use when you have a gene expression matrix (RNA-seq counts, microarray intensities, or normalized expression values) and need to perform gene set enrichment analysis on preranked gene lists. Use this skill when you want to rank genes by a univariate statistic (e.

Use when you have fit a linear model to normalized gene expression data (microarray intensities or RNA-seq counts) and need to test for differential expression across experimental conditions or contrasts.

Use when you have a large gene expression matrix (e.g., thousands of genes) from normalized microarray or RNA-seq data and need to reduce computational burden before running pathway enrichment analysis (e.g., GSEA or GESECA).

Use when when loading gene expression data (e.g., from GEO via getGEO or microarray ExpressionSet objects) that contains duplicate rows mapped to the same Gene ID, missing gene identifiers, or identifiers encoded with placeholder strings ('///') that indicate failed or ambiguous annotation.

Use when you have transcript-level quantification (salmon, Sailfish, or kallisto output) summarized to gene level by tximport, and you observe or suspect differential isoform usage across your experimental conditions.

Use when you have transcript-level quantification files (salmon quant.sf.gz, kallisto h5, or Sailfish output) and need to perform gene-level differential expression analysis.

Use when when you have a ranked list of gene-level statistics (e.

Use when you have a normalized gene expression matrix (log2-quantile normalized, filtered to high-variance genes) and a collection of annotated gene sets (e.g., Reactome, MSigDB pathways), and need to test whether specific pathways show significant coordinated expression shifts.

Use when your gene expression matrix or pathway collection uses identifier formats incompatible with your enrichment analysis tool (e.g., gene symbols vs. Entrez IDs), or you need to reconcile gene sets from multiple sources (Reactome, KEGG) that employ different naming conventions.

Use when you need to set up a local copy of a scientific software project (e.g., Scanpy) to run its test suite, modify source code, or prepare a feature or bugfix contribution. The canonical repository is hosted on GitHub and you do not have direct push access.

Use when you have spatial molecular data (e.g., coordinates from microscopy or sequencing assays stored in an AnnData object), you need to compute a k-nearest-neighbor graph for spatial statistics (e.

Use when after running fgsea() on a preranked gene list when you need to: (1) identify which pathways are most significantly enriched or depleted (lowest p-values), (2) distinguish between upregulated pathways (ES > 0) and downregulated pathways (ES < 0) within your gene set collection, (3).

Use when you have a GTF-formatted genome annotation file and need to generate alternative splicing events (exon skipping, intron retention, alternative splice sites, mutually exclusive exons) or transcript-isoform inclusion levels (PSIs).

Use when you have cloned a Python project (e.g., scverse/scanpy) that includes a hatch.toml configuration file and need to set up a consistent development or testing environment.

Use when after normalization (normalize_total, log1p transformation) and before PCA or other dimensionality reduction on raw or near-raw single-cell gene expression matrices.

Use when when you have completed DESeq2 differential expression analysis on RNA-seq count data and obtained p-values for each gene, use IHW if you want to improve power to detect true positives beyond standard independent filtering.

Use when you have a spatial transcriptomics dataset (AnnData object) with cell/spot coordinates and an associated tissue microscopy image file, and you need to compute image-derived features (e.

Use when analyzing RNA-seq count data from a DESeq2 workflow where you have fitted negative binomial generalized linear models and need to extract final results. Use it specifically when you want to identify genes with adjusted p-value below a predetermined significance threshold (e.g., α=0.

Use when you have completed DESeq differential expression analysis on a DESeqDataSet and obtained raw results with p-values across all genes.

Use when when comparing quantification results between two versions of a tool (e.g., salmon 2.0 Rust rewrite vs. C++ salmon 1.12.0), or when validating a new tool implementation against a reference version.

Use when when performing transcript- or gene-level differential expression analysis and your quantification tool (Salmon, Sailfish, or kallisto) has produced Gibbs sample or bootstrap sample replicates, and you want edgeR to account for inferential uncertainty rather than treating point estimates.

Use when when you have a GTF genome annotation and need to catalog all transcript isoforms and local alternative splicing event variants (exon skipping, intron retention, alternative splice sites, mutually exclusive exons, alternative first/last exons) before quantifying their inclusion levels.

Use when when performing gene-level differential expression analysis on RNA-seq data where samples may express different isoforms of the same gene at different relative abundances. Specifically: (1) you have transcript-level quantification (e.

Use when a Rust port or alternative implementation of a mapper (e.g., salmon, piscem) consistently maps 2–3% more reads than a C++ reference, especially on short reads.

Use when after PCA dimensionality reduction and scaling of normalized, log-transformed gene expression data, when you need to identify local cell neighborhoods (typically with k=15 neighbors) before applying clustering algorithms like Leiden or computing UMAP embeddings.

Use when your input is a spatial dataset (AnnData object with coordinate metadata) paired with a large tissue image, and you need to extract spatial features (via squidpy.im.calculate_image_features or similar operations) without loading the entire computation graph into memory.

Use when after constructing a kNN graph (via pp.neighbors) on preprocessed, scaled, and PCA-reduced single-cell expression data.

Use when apply Leiden clustering after constructing a k-nearest neighbor (kNN) graph from single-cell expression data when you need to partition cells into discrete, biologically meaningful clusters for downstream trajectory inference, differential expression testing, or graph abstraction.

Use when you have a normalized gene expression matrix (genes × samples) and an experimental design with known treatment groups or conditions, and you need to estimate the effect of those conditions on expression levels while accounting for sample-to-sample variability.