Machine Learning Skills

Predict RBP binding from RNA sequence using deep learning models (RBPNet sequence-to-signal, RNAProt RNN, GraphProt2 GCN with structure, DeepCLIP, DeepRiPe multi-modal CNN) for variant-effect prediction, in silico binding-site discovery, model interpretation, and transfer learning from CLIP and RBNS datasets. Use when computational prediction of RBP binding from sequence is needed, evaluating variant effects on binding without further wet-lab experiments, comparing model performance, or training a custom model on ENCODE eCLIP data.

Dimensionality reduction and clustering for single-cell RNA-seq using Seurat (R) and Scanpy (Python). Use for running PCA, computing neighbors, clustering with Leiden/Louvain algorithms, generating UMAP/tSNE embeddings, and visualizing clusters. Use when performing dimensionality reduction and clustering on single-cell data.

Infer orthologous genes and gene families across species using OrthoFinder3 (HOG-based phylogenetic orthology), SonicParanoid2, Broccoli, ProteinOrtho, OMA / FastOMA hierarchical orthologous groups, eggNOG-mapper, JustOrthologs, and TOGA whole-genome-alignment orthology. Use when building single-copy ortholog sets for phylogenomics, classifying co-orthologs and in/out-paralogs after gene duplication, propagating functional annotation via orthology with awareness of the ortholog conjecture, distinguishing speciation from duplication via gene-tree species-tree reconciliation, computing Quest-for-Orthologs benchmark performance, or running synteny-aware ortholog detection in WGD-affected lineages.

Analyzes CUT&RUN (Skene Henikoff 2017) and CUT&Tag (Kaya-Okur 2019) chromatin profiling data. Handles SEACR vs MACS2 peak calling (with the btaf375 2025 benchmark guidance), pA-MNase vs pA-Tn5 vs pAG-Tn5 chimera differences, E. coli spike-in carryover normalization, IgG-only control logic (no input), characteristic fragment-size signatures (25-75 bp for CUT&Tag), and lower depth requirements (5M reads typical vs 25M for ChIP). Use when calling peaks from CUT&RUN/CUT&Tag, scaling by E. coli spike-in carryover, choosing SEACR norm mode, or comparing CUT&RUN/Tag results to traditional ChIP.

Preprocessing and harmonization of multi-omics data before integration. Covers normalization, batch correction, feature alignment, and missing value handling across data types. Use when preparing multi-omics datasets for integration analysis.

Sequence-based deep learning for ATAC-seq using chromBPNet, BPNet, scBasset, or EnFormer. Use when correcting Tn5 bias with neural networks beyond k-mer models, predicting per-base accessibility profiles, scoring in silico variant effects at GWAS or rare-variant SNPs, discovering motifs via DeepLIFT/TF-MoDISco from a trained model, or generating cell-type-specific accessibility predictions for unobserved cell states.

Produce and interpret PCA, t-SNE, UMAP, and PHATE plots for high-dimensional omics data with rigor about which method preserves what (variance, local structure, manifold, transitions), hyperparameter sensitivity, and the well-documented limits of 2D embeddings. Covers PCA biplot/scree/loadings, t-SNE PCA initialization (Kobak-Berens 2019), UMAP n_neighbors/min_dist trade-offs, and the Chari-Pachter 2023 critique. Use when visualizing high-dimensional data — bulk PCA, single-cell embeddings, multi-omics integration projections.

Predict protein-coding genes in eukaryotic genomes using BRAKER3 for combined RNA-seq and protein evidence, or GALBA for protein-only evidence. Runs Augustus with trained parameters for accurate gene models. Use when annotating a newly assembled eukaryotic genome or improving existing gene predictions.

Query and download from NCBI Gene Expression Omnibus (GEO) and EMBL-EBI's BioStudies/ArrayExpress mirror. Use when finding expression datasets, navigating SuperSeries vs SubSeries, choosing between series-matrix (submitter-normalized) and raw supplementary files, downloading via GEOparse (Python) or GEOquery (R/Bioconductor), linking GEO to SRA for raw reads, or distinguishing GSE/GSM/GPL/GDS record types. Encodes the SuperSeries trap, the series-matrix normalization-trust caveat, GEOmetadb deprecation, ArrayExpress migration to BioStudies, and processed-vs-raw decision matrix.

End-to-end gene regulatory network inference pipeline from processed single-cell data to regulon discovery and perturbation simulation. Supports RNA-only (pySCENIC) and multiome (SCENIC+) paths. Use when building gene regulatory networks from single-cell transcriptomic or multiome data.

Interactive cell type annotation for IMC data. Covers napari-based annotation, marker-guided labeling, training data generation, and annotation validation. Use when manually annotating cell types for training classifiers or validating automated phenotyping results.

Build publication-quality figures with matplotlib using the object-oriented Figure/Axes API, constrained_layout, rcParams customization, TrueType (Type-42) font embedding for journal submission, and CVD-safe palettes. Covers seaborn integration, common chart types, axis formatting, and the small gotchas that distinguish reproducible matplotlib from notebook scratch. Use when producing publication figures in Python — RNA-seq scatter, single-cell embeddings, generic biological plotting.

Polish assemblies and call variants from Oxford Nanopore data using medaka. Uses neural networks trained on specific basecaller versions. Use when improving ONT-only assemblies or calling variants from Nanopore data without short-read polishing.

Analyze metabolite-mediated cell-cell communication using MeboCost for metabolic signaling inference between cell types. Predict metabolite secretion and sensing patterns from scRNA-seq data. Use when studying metabolic crosstalk between cell populations or metabolite-receptor interactions.

Predict peptide-MHC class I and II binding affinity using MHCflurry and NetMHCpan neural network models. Identify potential T-cell epitopes from protein sequences. Use when predicting MHC binding for vaccine design or neoantigen identification.

Implements nested cross-validation and stratified splits for unbiased model evaluation on biomedical datasets. Prevents data leakage and overfitting in biomarker discovery. Use when validating classifiers or optimizing hyperparameters on omics data.

Build maximum likelihood phylogenetic trees using IQ-TREE2 and RAxML-NG with expert model selection, branch support assessment, and topology testing. Use when inferring publication-quality ML trees, selecting substitution models, interpreting bootstrap and concordance factor support, or running partitioned phylogenomic analyses.

Standardizes molecular structures using ChEMBL chembl_structure_pipeline and RDKit rdMolStandardize covering sanitization, salt/solvent stripping, neutralization, tautomer canonicalization, stereochemistry standardization, mixture handling, and isotope normalization. Explicitly compares ChEMBL pipeline, canSARchem, and PubChem standardization choices. Use when preparing libraries for QSAR training, joining datasets across sources, deduplicating compound collections, or building canonical compound registries.

End-to-end multi-omics integration workflow. Orchestrates data harmonization, MOFA/mixOmics integration, factor interpretation, and downstream analysis across transcriptomics, proteomics, metabolomics, and other modalities. Use when integrating multiple omics datasets.

Perform multiple sequence alignment using MAFFT, MUSCLE5, ClustalOmega, or T-Coffee. Guides tool and algorithm selection based on dataset size, sequence divergence, and downstream application. Use when aligning three or more homologous sequences for phylogenetics, conservation analysis, or evolutionary studies.

Download genome assemblies, gene records, and ortholog data from NCBI using the modern Datasets v2 CLI (replaces assembly_summary.txt scraping and many EFetch workflows). Use when bulk-pulling genome assemblies, gene metadata across species, ortholog sets, or BLAST databases; when E-utilities are too slow for genome-scale work; or when automatic checksum verification, parallel download, and clean accession-driven retrieval are required. Encodes the JSON-lines output format, dataformat conversion, --dehydrated for cloud workflows, and when Datasets is/isn't the right tool.

End-to-end outbreak investigation from pathogen isolates to transmission networks. Orchestrates MLST typing, AMR surveillance, phylodynamic dating, and transmission inference with TransPhylo. Use when investigating disease outbreaks or tracking pathogen transmission chains.

Simulate transcription factor perturbation effects on cell state using CellOracle, which integrates GRN inference with in silico knockout and overexpression modeling. Predicts cell identity shifts and differentiation trajectory changes from TF perturbations. Use when predicting the effect of transcription factor knockouts, planning perturbation experiments, or identifying driver TFs for cell fate transitions.

Detect positive (diversifying / episodic / pervasive) selection using codon dN/dS frameworks. Implements PAML codeml site models (M0/M1a/M2a/M7/M8/M8a), branch models, branch-site model A (Zhang 2005), and HyPhy methods (BUSTED, BUSTED-S, BUSTED-MH, BUSTED-PH, MEME, FEL, FUBAR, aBSREL, SLAC, RELAX, GARD, FUBAR-MH). Includes McDonald-Kreitman framework (asymptotic alpha, impMKT, polyDFE, DFE-alpha, GRAPES) for within-species + divergence inference, RERconverge for trait-correlated rate shifts, CSUBST for convergent substitution, and PhyloAcc for accelerated noncoding evolution. Use when testing adaptive evolution at codons, branches, or full gene; running GARD recombination pre-screen; controlling alignment-error and gBGC false positives; reconciling PAML vs HyPhy results; or performing genome-scale selection scans.

Protein grouping and inference from peptide identifications. Use when resolving protein ambiguity from shared peptides. Handles protein groups and protein-level FDR control using parsimony and probabilistic approaches.

Work with sparse matrices for memory-efficient storage of count data. Use when dealing with single-cell data or large bulk RNA-seq datasets where most values are zero.

Detect syntenic blocks and structural rearrangements between genomes using MCScanX (Wang 2012), JCVI/MCScan (Tang 2008 Python), GENESPACE (Lovell 2022) for orthology-anchored riparian visualization, SyRI for structural variation, AnchorWave for sequence-level synteny, i-ADHoRe 3.0 for highly diverged species, SynNet for synteny networks, and ntSynt for multi-genome macrosynteny. Use when identifying collinear gene blocks across species, distinguishing macrosynteny from microsynteny, detecting inversions/translocations/duplications, anchoring orthology in WGD lineages, producing publication riparian plots, computing synteny block age via Ks (cross-references whole-genome-duplication), or running synteny-aware ortholog inference in polyploids.

Infer developmental trajectories and pseudotime from single-cell RNA-seq data using Monocle3, Slingshot, and scVelo for RNA velocity analysis. Use when inferring developmental trajectories or pseudotime.

Models continuous temporal trajectories from bulk or time-resolved omics data using generalized additive models (mgcv), spline regression, and changepoint detection (segmented, ruptures). Fits smooth gene expression curves and tests trajectory differences between conditions. Use when fitting non-linear temporal models to bulk time-series data or comparing developmental trajectories across conditions. Not for single-cell pseudotime (see single-cell/trajectory-inference).

Infer pathogen transmission networks and identify likely transmission pairs using TransPhylo and outbreak reconstruction algorithms. Estimate who-infected-whom from genomic and epidemiological data. Use when investigating outbreak transmission chains or identifying superspreaders.

> Biohub ESMFold2 / ESMFold2-Fast all-atom co-folding (Candido et al. 2026, github.com/Biohub/esm). Single-sequence and MSA modes; protein, DNA, RNA, ligand (CCD/SMILES), modified residues. FoldBench Ab-Ag 50-55%, PPI 70-77% DockQ-pass. Also covers the ESMC-{300M,600M,6B} protein language models from prediction, and the SAE interpretability head. MIT-licensed weights on structures with single-sequence input, (2) Validating designed binders with ESMFold2-Fast, (3) Running ESMFold2 with MSA input, (4) Getting ESMC embeddings or per-residue mutation scores, (5) Choosing kernel backend and sampling-step settings for paper-faithful throughput.

> Score, embed, and generate DNA sequences with Evo 2, a long-context genomic (1) Computing per-nucleotide or per-sequence likelihoods for variant effect scoring, (2) Embedding genomic windows for downstream classification, (3) Generating DNA conditioned on a prefix, (4) Scoring regulatory or coding regions across species.

Register a model service in the managed family — a local model server container the daemon starts/stops on demand, or a remote upstream model API (https). Read the runbook, allocate a port (local only), compose idempotent start/stop scripts (local only), register once. Load when the user wants a model service available for inference, or when list_compute shows managed endpoints.

> Structure prediction using OpenFold3, an open-weights PyTorch reproduction of AlphaFold3 from the AlQuraishi Lab. Use this skill when predicting protein/nucleic-acid/ligand complex structures with an Apache-2.0-licensed AF3 reimplementation.

> Embed and annotate single-cell expression data with scGPT, a foundation model (1) Producing cell embeddings from an AnnData for clustering/integration, (2) Zero-shot or fine-tuned cell-type annotation, (3) Gene-level representation for perturbation/GRN tasks. For probabilistic single-cell models (scVI etc.), use the scvi-tools library.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Call a registered model endpoint over its native HTTP API from the endpoint's scoped inference kernel (BASE_URL preloaded). Load once a task needs predictions from a registered model endpoint.

Semantic Similarity Index for disease research literature using PubMedBERT embeddings

| Fetch a region of cis-eQTL summary statistics from EBI eQTL Catalogue v7+ via tabix-on-FTP. Use when an agent needs eQTL beta / SE / p-value for every variant in a window around a gene's TSS for one specific dataset

>- Download genomes, genes, virus sequences, and taxonomy data from NCBI using the datasets and dataformat CLI tools.

Local scVI/scANVI-based single-cell latent embedding and batch-aware integration from raw-count .h5ad or 10x Matrix Market input, with stable integrated AnnData export for downstream latent analysis.

Local Scanpy pipeline for single-cell RNA-seq QC, optional doublet detection, clustering, marker discovery, optional CellTypist annotation, optional latent downstream mode from integrated.h5ad/X_scvi, and optional dataset-level plus within-cluster contrastive marker analysis from raw-count .h5ad or 10x Matrix Market input.

> Query the DILIrank/FDA Liver Toxicity Knowledge Base (LTKB). Use whenever the user asks about drug-induced liver injury (DILI) risk, hepatotoxicity classification, or wants to look up any drug (by name, LTKB ID, or DILIst ID) in the DILIrank or DILIst datasets.

> Query the DrugLib.com Drug Review Dataset (UCI #461). Use whenever the user asks about patient drug reviews, drug effectiveness ratings, side-effect profiles, or condition-specific treatment experiences from DrugLib.com.

> Query the NCI-60 Molecular Target (Protein) database from the Developmental Therapeutics Program. Use when the user asks about protein expression of drug targets across the NCI-60 cancer cell line panel, or wants to look up a gene, cell line, or cancer panel in the NCI DTP molecular target dataset.

> Query the WebMD Drug Reviews dataset (~362 k patient reviews, 2007–2020). Use whenever the user asks about patient-reported drug effectiveness, ease of use, satisfaction ratings, side effects, or reviews for a specific drug or medical condition.

Manages persistent research memory across ideation and experimentation cycles. Maintains two stores: Ideation Memory M_I (feasible/unsuccessful directions) and Experimentation Memory M_E (reusable strategies for data processing, model training, architecture, debugging). Three evolution mechanisms: IDE (after research-ideation), IVE (after experiment failure — classifies failures as implementation vs fundamental), ESE (after experiment success — extracts reusable strategies). Use when: updating memory after completing research-ideation cycles or experiment pipelines, classifying why a method failed (implementation vs fundamental failure), starting a new research cycle needing prior knowledge, user mentions 'update memory', 'classify failure', 'what worked before', 'research history', 'evolution'. Do NOT use for running experiments (use experiment-pipeline), debugging experiment code (use experiment-craft), or generating ideas (use research-ideation).