LLMs-Universal-Life-Science-and-Clinical-Skills-

Operate MedSAM2 for promptable segmentation of 3D medical images and medical videos, including CT lesion propagation, MRI volumes, RECIST-guided prompts, efficient CPU-oriented variants, training, and 3D Slicer integration. Use when generating or validating volumetric masks from sparse prompts or propagating masks through image slices or video frames.

Build reproducible healthcare imaging pipelines with Project MONAI for DICOM, NIfTI, pathology, and multidimensional imaging tasks including preprocessing, augmentation, training, sliding-window inference, evaluation, model bundles, labeling, and deployment. Use when implementing medical image classification, segmentation, registration, detection, generative, or foundation-model workflows in PyTorch.

Operate Google TxGemma prediction and chat models for therapeutic property prediction across small molecules, proteins, nucleic acids, diseases, targets, and cell lines. Use when formatting Therapeutics Data Commons tasks, choosing TxGemma model size or variant, running local or Model Garden inference, fine-tuning on private therapeutic data, or evaluating TxGemma in drug-discovery workflows.

Evaluate and operate released Profluent OpenCRISPR gene-editing systems, especially OpenCRISPR-1, for controlled research workflows using its published Cas9-like protein, compatible guide RNA designs, protocols, licensing, specificity testing, and experimental validation. Use when comparing OpenCRISPR-1 with SpCas9, planning nonclinical editing studies, or assessing use in nuclease, nickase, deactivated, base, prime, or epigenome-editing contexts.

Operate CZI TranscriptFormer cross-species generative single-cell models to produce cell embeddings, contextual gene embeddings, likelihoods, zero-shot classifiers, disease-state representations, and regulatory analyses from raw-count AnnData files. Use when selecting TF-Sapiens, TF-Exemplar, or TF-Metazoa, processing in- or out-of-distribution species, or scaling embedding extraction across GPUs.

Operate ByteDance Protenix-v2 for open biomolecular structure prediction of proteins, antibodies, nucleic acids, ligands, and complexes using JSON inputs, MSA and template features, constraints, and inference-time sampling. Use when running Protenix locally or through its server, comparing AlphaFold3-style open models, or building reproducible co-folding evaluations.

Build and evaluate medical text and vision applications with Google MedGemma, including MedGemma 1.5 workflows for CT, MRI, whole-slide pathology, longitudinal chest X-rays, lab reports, and EHR text. Use when prototyping, fine-tuning, deploying, or validating MedGemma-based health AI under clinical data and safety controls.

Run Boltz-2 biomolecular interaction predictions for protein, nucleic-acid, ligand, and complex structures with binding-affinity outputs. Use for hit discovery, binder-versus-decoy prioritization, hit-to-lead comparisons, lead optimization, complex modeling, or reproducible Boltz YAML and batch inference workflows.

Use Google DeepMind AlphaGenome to predict tissue-aware regulatory effects of DNA sequence variants across expression, splicing, chromatin, and contact-map outputs. Use when prioritizing noncoding variants, comparing reference and alternate alleles, visualizing predicted regulatory changes, or designing focused AlphaGenome API analyses.

Operate Arc Institute Evo 2 for long-context DNA sequence scoring, zero-shot variant effect analysis, genomic embeddings, sequence generation, and model deployment. Use when a task explicitly needs Evo 2, million-base genomic context, DNA likelihood comparisons, genomic foundation-model embeddings, or generated DNA candidates across species.

Operate Microsoft BioEmu to sample approximate equilibrium conformational ensembles for protein monomers from amino-acid sequences or supplied MSAs. Use when studying protein flexibility, alternative conformations, free-energy landscapes, disorder, ensemble generation, physical steering, or downstream side-chain reconstruction and MD relaxation.

Reference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Use when executing implementation plans with independent tasks in the current session

Use when you have a spec or requirements for a multi-step task, before touching code

Use when creating new skills, editing existing skills, or verifying skills work before deployment

Load when removing batch effects from a multi-cohort bulk RNA-seq dataset using ComBat (R or Python implementation). Skip if there is only one batch, or for single-cell batch integration (use sc-batch-integration), or for spatial multi-slice integration (use spatial-integrate).

Load when discovering gene co-expression modules and hub genes in a bulk RNA-seq cohort via WGCNA-style soft-thresholded networks. Skip for direct DE comparison (use bulkrna-de) or PPI lookup of an existing gene list (use bulkrna-ppi-network); single-cell co-expression uses sc-grn instead.

Load when comparing gene expression between two conditions in bulk RNA-seq count data. Skip when the data is single-cell (use sc-de) or spatial (use spatial-de), or when you need exon-level alternative splicing (use bulkrna-splicing).

Load when estimating cell-type proportions in bulk RNA-seq samples from a single-cell or signature-matrix reference. Skip if the data is already single-cell (no deconvolution needed) or for spatial deconvolution (use spatial-deconv).

Load when running pathway / GO term enrichment on a bulk RNA-seq DE result list. Skip if the input is single-cell (use sc-enrichment), spatial (use spatial-enrichment), or for metabolite pathways (use metabolomics-pathway-enrichment).

Load when converting gene identifiers between Ensembl, Entrez, and HGNC symbol in a bulk RNA-seq count matrix. Skip if the input is already in the desired identifier system, for organisms outside human/mouse, or for non-bulk-counts inputs.

Load when querying STRING for the protein-protein interaction subgraph induced by a bulk RNA-seq DEG list and finding hub genes. Skip for pathway enrichment of the same list (use bulkrna-enrichment) or for de novo co-expression network discovery (use bulkrna-coexpression).

Load when checking a bulk RNA-seq count matrix for library-size outliers, gene detection rates, and sample-sample correlation before DE. Skip if data is raw FASTQ (use bulkrna-read-qc) or aligner logs (use bulkrna-read-alignment), or for single-cell counts (use sc-qc).

Load when summarising STAR / HISAT2 / Salmon alignment-rate logs in bulk RNA-seq. Skip if data is raw FASTQ (use bulkrna-read-qc) or already counted (use bulkrna-qc), or for genome-DNA alignment (use genomics-alignment).

Load when checking raw FASTQ quality (Phred / GC / adapter / Q20-Q30) before alignment in bulk RNA-seq. Skip if reads are already aligned (use bulkrna-read-alignment) or counted (use bulkrna-qc), or for single-cell FASTQ (use sc-fastq-qc).

Load when summarising rMATS / SUPPA2 alternative-splicing output and identifying significant differential splicing events. Skip if you only have count-level DE (use bulkrna-de) or for splicing in single-cell or spatial data (currently unsupported).

Load when stratifying patients by gene expression and testing for survival differences (Kaplan-Meier + Cox) in bulk RNA-seq. Skip if no time-to-event clinical data exists, or for non-bulk cohorts (single-cell / spatial survival is not supported).

Load when placing bulk RNA-seq samples on a single-cell reference's pseudotime axis (NNLS deconvolution + nearest-neighbour mapping). Skip for plain cell-type proportions (use bulkrna-deconvolution alone) or for native single-cell trajectory inference (use sc-pseudotime).

Load when computing alignment QC metrics (mapping rate, MAPQ distribution, insert size, duplicate rate, proper-pair rate) from a SAM or BAM file produced by any short-/long-read aligner (BWA / Bowtie2 / Minimap2). Skip when running the alignment step itself or when only FASTQ-level QC is needed (use `genomics-qc`).

Load when computing genome-assembly QC metrics — N50/N90, L50/L90, total length, contig count, GC content, longest-contig — from a FASTA produced by any assembler (SPAdes / Megahit / Flye / Canu). Skip when running the assembly itself or when assessing alignment quality (use `genomics-alignment`).

Load when calling CNV segments via CBS-style segmentation on a bin-level log2-ratio CSV from exome / WGS coverage — emits per-segment 5-class CN state (`amplification` / `gain` / `neutral` / `loss` / `deep_deletion`), per-chromosome summary, genome-fraction-altered. Skip when working with single-cell / spatial CNV (use `spatial-cnv`).

Load when summarising a peak file (BED / narrowPeak) from ATAC-seq / ChIP-seq / CUT&Tag — peak count, width distribution, per-chromosome counts, score statistics. Skip when calling peaks from BAM (run MACS / Genrich externally first) or when working with single-cell ATAC (use `scatac-preprocessing`).

Load when summarising a phased VCF (output of WhatsHap / SHAPEIT5 / Eagle2) — phased fraction of het variants, phase-block N50, PS-field parsing, pipe-delimited genotype detection. Skip when the input is unphased (run a phaser first) or when calling small variants (use `genomics-variant-calling`).

Load when running pre-alignment FASTQ quality control — Phred quality scores, Q20/Q30 rates, GC / N content, read-length distribution, adapter-contamination detection. Skip when working with already-aligned BAMs (use `genomics-alignment`) or when peak / variant files are the input (use the relevant downstream skill).

Load when summarising structural variants from an SV VCF (DEL / DUP / INV / TRA) — BND-notation parsing, size classification, per-type counts. Skip when working with small SNVs / indels (use `genomics-variant-calling`) or calling SVs from BAM (run Manta / Delly / Sniffles first).

Load when summarising functional impact of an annotated variant CSV — per-IMPACT counts (HIGH / MODERATE / LOW / MODIFIER), top consequences, gene-affected count. Skip when input is a raw VCF (convert with `bcftools +split-vep` first), when calling raw variants (use `genomics-variant-calling`), or filtering VCFs (use `genomics-vcf-operations`).

Load when summarising small variants (SNVs / indels) from a VCF or computing demo-pattern variant statistics (Ti/Tv ratio, per-chromosome distribution, SNP / indel split). Skip when filtering / merging VCFs (use `genomics-vcf-operations`), when calling structural variants (use `genomics-sv-detection`), or when adding functional annotations (use `genomics-variant-annotation`).

Load when summarising / filtering a VCF — variant classification (SNP / MNP / INS / DEL / COMPLEX), Ti/Tv ratio, QUAL / DP threshold filtering, INFO-field parsing. Skip when the input is a BAM (use `genomics-variant-calling` upstream first) or when adding functional annotations (use `genomics-variant-annotation`).

Load when extracting GEO accessions, dataset metadata, and downloadable references from a scientific paper (PDF / URL / DOI / PubMed ID / raw text) for downstream omics analysis. Skip when the dataset is already in hand or when only routing a query (use `orchestrator`).