Claude Skills by HolobiomicsLab

Use when when you need to compare spectrum prediction models fairly across different encoder architectures (GNN vs. FFN vs. Transformer), and you require equivalent settings (same covariates, identical hyperparameter sweeps) to isolate the effect of the encoder design.

Use when you have molecular structures (SMILES or SDF format) that need to be matched against MS/MS spectra, or you need to compute similarity between query spectra and a reference library of compounds.

Use when you have statistically significant LC-MS features grouped into structural clusters (isotopologue groups, adduct groups, cross-assay links) and correlation cluster assignments from upstream MamsiStructSearch, and you need to create an interactive graph representation suitable for Cytoscape.

Use when your XCMS-processed LC-MS dataset exhibits retention-time drift or misalignment artifacts—particularly when analyzing hundreds of samples, data acquisition spans longer than one week, or you observe feature groups with inconsistent m/z or RT that XCMS grouped under a single global warping.

Use when you have raw or peak-picked mass spectrometry data in HDF5 format that needs to be loaded into memory for downstream processing (feature alignment, isotope detection, CCS calibration), or when you need to export annotated feature tables with isotopologue metadata back to HDF5 for archival.

Use when you have multidimensional MS data converted to MZA HDF5 format (from Agilent .d, Bruker .d with ion mobility, Thermo .

Use when after invoking the DEIMoS CLI with a configuration file and allowing the Snakemake workflow to execute, use this skill to confirm successful completion of all workflow rules.

Use when after training a DeepMSProfiler model and generating per-sample predictions: when you need to display Pearson or Spearman correlation coefficients between individual metabolite signals and disease class labels in a matrix form suitable for publication or exploratory review of.

Use when your peptide sample has been enriched with stable isotope labels (13C, 15N, etc.) at a known incorporation probability (e.g., 52% 13C).

Use when when building a Graph Transformer model for continuous property prediction on molecules with associated experimental or instrumental metadata (e.g., retention time prediction across different chromatographic columns, methods, or conditions).

Use when you have a pre-computed hierarchical dendrogram from correlation-based clustering of LC-MS features (with fixed linkage criterion and distance metric) and need to decide whether a single constant-threshold cut or data-driven silhouette optimization better resolves the underlying cluster.

Use when when you have a large combinatorial chemical space (e.g., all regioisomeric positions of lipid double bonds across carbon counts and saturation levels) and need to query it repeatedly against experimental mass spectrometry spectra.

Use when when building a reference library for high-throughput spectral matching against experimental MS/MS data, and you need to support millions to billions of queries per second on a standardized dataset.

Use when immediately after parsing and validating raw LC-MS/MS data files (mzML, mzXML, or vendor formats) when you need to prepare spectral data for fragmentation tree computation, isotope pattern analysis, or molecular formula ranking within the SIRIUS framework.

Use when when you have millions of high-dimensional objects (e.g., MS/MS spectra converted to feature-hashed vectors) and need to compute pairwise similarities or retrieve nearest neighbors efficiently.

Use when you have a pre-trained MS/MS spectral embedding model and need to validate that it achieves strong and consistent retrieval performance on curated spectral libraries that represent real-world data quality standards.

Use when you have acquired high-resolution MS/MS spectra in mzML, mzXML, or MGF format and need to prepare them for large-scale clustering or similarity searching.

Use when you have centroided LC- or GC-HRMS data (in mzML format, ideally from data-dependent acquisition) and need to identify potential PFAS candidates from a large feature list.

Use when when you have high-resolution tandem mass spectra from proteomics experiments and need to search against a spectral library for both unmodified peptides and those with unknown or variable post-translational modifications.

Use when you have high-resolution MS2 data in .ms2 format from lipid A samples and need to perform automated structure annotation to identify lipid A molecular variants and their fragmentation patterns at scale.

Use when when you have MGF-formatted mass spectrometry peak files (typically from proteomics experiments) numbering in the millions of spectra that require grouping into similar MS/MS clusters, and runtime must be kept under 15 minutes rather than hours.

Use when when you have a trained MS/MS spectral embedding model and need to measure compound identification accuracy on a held-out test set, but want to mitigate sensitivity to a single random train/test split. Use this skill if the original training set split is fixed (e.

Use when when you need to establish a reproducible inventory of compounds for LC-MS/MS simulation studies, particularly to determine how many unique molecular formulas fall within a target m/z window (e.g., 100–1000 Da) and MS1 detection level.

Use when when you need to create in silico LC-MS/MS experiments with diverse chemical backgrounds for testing fragmentation strategies or acquisition controllers, and you want the chemical diversity to reflect real metabolomic samples. Use this when you have a target m/z range (e.

Use when you have downloaded raw HMDB data (hmdb_metabolites.zip or pickle file) and need to generate a representative set of chemical objects for simulating LC-MS/MS acquisition strategies.

Use when you have identified one or more proton NMR spectral regions-of-interest (ROIs)—defined by lower and upper chemical-shift bounds in ppm—from complex biological samples (serum, saliva, urine, tissue, CSF) and need to generate a ranked list of plausible metabolite identities.

Use when you have a feature list (m/z, retention time, intensity) from HRMS data and want to identify potential PFAS homologous series for prioritization.

Use when after data normalization (Step 7) on the preprocessed feature matrix when you need to identify and flag anomalous samples before statistical testing or biomarker discovery.

Use when when you have raw HPLC column specifications from RepoRT or similar metadata repositories and need to prepare them as input features for machine learning models. Apply this skill before featurizing molecular structures or training graph transformers for retention time prediction.

Use when you have raw or processed HRMS/MS data from Q-Exactive, Agilent Q-TOF, Bruker Q-TOF, or SCIEX Q-TOF instruments in formats such as mzML, CSV peaklists, or vendor-specific formats, and you need to extract experimental fragment m/z values and their intensities for comparison against.

Use when you have LC- or GC-HRMS data in mzML format and a feature list (CSV/TSL/Excel) from external feature detection software (e.

Use when after MS-CleanR has filtered and clustered LC-MS features and formatted them for MS-FINDER input (m/z, retention time, MS/MS spectra).

Use when when clustering large-scale mass spectrometry datasets (millions of MS/MS spectra in MGF format) where runtime is a bottleneck and you have access to NVIDIA GPU resources (GTX 3090 for large datasets like PXD000561, or GTX 1080Ti for smaller datasets).

Use when when implementing multiple competing model architectures (e.

Use when when training a fresh NeatMS CNN model from scratch on LCMS peak classification and you need to determine which optimizer (Adam vs. SGD) and learning rate will produce an AUC ROC > 0.9 without overfitting.

Use when you receive mass spectrometry data through heterogeneous identifier formats—specifically when the input could be a GNPS Task ID, a Universal Spectrum Identifier (USI), or a Feature-Based Molecular Networking (FBMN) identifier—and you need to programmatically determine which format was.

Use when you have experimental MS/MS spectra and need to assign definitive molecular identities by matching against a curated spectral library.

Use when you have a two-dimensional MS map (m/z vs retention time) from GC–MS or LC–MS data and need to discriminate analytes and identify marker features without false positives from peak picking; particularly useful for untargeted metabolomics at ppb sensitivity (e.

Use when you have raw mass spectrometry data files (mzML, NetCDF, or vendor formats) with unknown or mixed acquisition modalities, and you need to automatically determine whether the input is LC-MS, GC-MS, IMS (ion mobility spectrometry), or MS imaging (e.

Use when you have a metabolomics dataset with left-censored missing values (e.g., below limit of quantification in LC/MS or GC/MS) and need to evaluate multiple imputation approaches.

Use when you have acquired imaging mass spectrometry (IMS) data stored in imzML format (accompanied by an .ibd ion binary data file) and need to load it into a Python-based spatial metabolomics workflow.

Use when you are repeatedly querying or iterating over multidimensional MS data stored in MZA HDF5 format (retention time, drift time, m/z dimensions) and profiling shows that repeated disk I/O for the same metadata or scan ranges dominates runtime.

Use when after feature filtering and clustering have been completed in MS-CleanR and you have a cleaned feature set with m/z, retention time, and MS/MS spectra data ready for structural assignment. Use it when you need to identify unknown compounds by querying multiple chemical databases (e.

Use when you have a collection of compound structures in SDF format (e.g., DNA adduct structures) and need to systematically generate predicted fragment spectra across a defined ionization level and mass range to populate a reference spectral database or validate experimental fragmentation patterns.

Use when you have candidate metabolite structures (from database lookup or enumeration) and experimental MS/MS spectra (mzML, mzXML format), and need to rank candidates by how well their predicted fragments match observed peaks.

Use when when you have a list of SMILES strings representing chemical structures and need to create paired SMILES-spectrum training data for a generative model (like MSGO) without requiring experimental mass spectra.

Use when you have an LCMS feature table (from XCMS, MS-DIAL, MZmine2, or other feature extraction software) with m/z, retention time, and intensity columns, plus MS2 spectral annotations from DDA data, and you need to identify which features are in-source fragments rather than distinct metabolites.

Use when you have an annotated MS/MS spectral dataset with structure metadata (InChI or SMILES records) linked to InChIKey identifiers, and you observe that some InChIKeys are associated with multiple or variant InChI strings due to curation inconsistencies, stereoisomerism notation differences, or.

Use when when you have raw MS/MS spectra from repositories like GNPS that lack or have inconsistent chemical structure annotations (InChI/SMILES), and you need to produce a curated dataset with uniform 14-character InChIKey and SMILES/InChI annotations for downstream machine learning or similarity.

Use when you have a ranked list of library candidates (top 2000 by MS2Deepscore) from MS/MS spectral matching and need to re-rank them using structural metadata to distinguish true analogues and exact matches from false positives.