Claude Skills by HolobiomicsLab

Use when when you have real LC-MS/MS experimental data in mzML format (e.

Use when you have raw LC- or GC-HRMS data from vendor instruments (ESI or APCI ionization) that needs to be converted to a vendor-neutral format for non-target screening, or you already have mzML files that require loading into a Python environment for downstream feature detection and MS2 spectral.

Use when after running an Environment simulation in ViMMS that has generated MS1 and/or MS/MS scans from a virtual mass spectrometer and controller pair. Use this skill when you need to preserve the generated scans in a standard format compatible with existing metabolomics software (e.

Use when after running a ViMMS simulation loop with a fragmentation controller (e.

Use when you have raw LC-HRMS metabolomics data in mzML or ABF format that needs to be processed through a reproducible pipeline. Use this skill when: (1) you have public or proprietary .mzML LC-MS datasets (e.g. from MetaboLights, MassIVE, or PRIDE);

Use when you have one or more mzML files (XML-based mass spectrometry data) and need to convert them into mzPeak format for downstream analysis, archival, or integration with tools that consume Parquet-based spectra.

Use when you have raw mass spectrometry data (vendor formats, mzML, or existing mzPeak files) and need to: (1) convert to mzPeak format for long-term storage and interoperability across languages and tools; (2) load mzPeak spectrum or chromatogram data into memory as structured tables for analysis;

Use when you have an mzPeak file (uncompressed ZIP archive containing Parquet files) and need to extract and work with spectrum metadata (scan descriptions, precursors, selected ions), spectrum signal data (profile or centroid m/z and intensity arrays), or chromatogram data.

Use when you have a set of natural product molecules (or suspected natural products) in SMILES, InChI, or SDF format and need a chemical representation suitable for biosynthetic classification, structural screening, or machine learning tasks where standard generic fingerprints (e.

Use when after curating and integrating structure-organism pairs from multiple source databases, and before publishing or using the dataset for computational research. Apply this skill when you have aggregated organism counts binned by structural diversity (e.

Use when you have LC-MS/MS DDA metabolomics data (positive and/or negative ionization modes) and sample metadata (originating taxon) for one or more samples, and you need to generate a Wikidata-connected RDF knowledge graph for integrated natural products analysis, taxonomy-aware compound.

Use when processing LC-MS metabolomics studies with >10 samples where sample count and memory constraints make pairwise mass alignment infeasible.

Use when you have millions of MS/MS spectra to cluster and have already constructed nearest neighbor indexes (partitioned Voronoi diagrams of spectrum vectors bucketed by precursor m/z).

Use when you have negative-mode MS/MS spectra with annotated molecular formulas and negative adducts (from repositories like MassIVE or MetaboLights), and your current formula inference model is restricted to positive mode only.

Use when after library-matching has produced ranked candidate spectra with MS2Deepscore embeddings.

Use when you have loaded raw Agilent Unknowns Analysis CSV output with required columns (Component.RT, Base.Peak.MZ, Component.Area, Compound.Name, Match.Factor, File.

Use when you have a GNPS GraphML molecular network and need to isolate cohesive subsets of spectra (components) before analyzing which fragmentation patterns explain them.

Use when after biclustering a normalized microbe-metabolite feature attribution score matrix into distinct functional modules, use this skill to summarize pairwise module interactions by aggregating scores between all microbe-metabolite pairs from different modules.

Use when after executing a molecular networking workflow on GC-MS data that has been processed through auto-deconvolution, and a published reference network exists from a prior analysis of the same or analogous dataset.

Use when after retrieving a molecular network file (GraphML or JSON format) from GNPS_GC following submission of deconvolved GC-MS spectra. Use this skill to confirm the network structure is sound before performing chemical similarity searches, community detection, or annotation.

Use when after identifying statistically significant features and assigning them to structural clusters (isotopologue groups, adduct groups, cross-assay links) and correlation clusters via MamsiStructSearch.

Use when you have raw mzML files and feature tables (CSV from mzMine or XCMS) for LCMS data, have generated training/validation/test batches with known class imbalance, and need to train a CNN model from scratch to achieve AUC ROC > 0.9 for distinguishing true from false positive MS1 peaks.

Use when you have a working base MPNN model (e.g., chemprop) and need to add task-specific feature processing layers (spectral, electronic, or domain features) to improve predictions on a specialized molecular property or spectrum.

Use when you have MS/MS spectra with unknown precursor m/z values and need to assign the most likely chemical formula and adduct type (e.g., [M+H]+, [M+Na]+, [M+K]+) in a de novo setting where spectrum database matching is unavailable or undesirable.

Use when when you have a pre-trained encoder (e.g., TCN spectrum encoder in FIDDLE) that has learned useful representations on a source task (e.g., MS/MS spectrum encoding), and you want to train lightweight task-specific modules (e.

Use when when training a neural network to predict metabolite abundances from microbiome data (or similar paired multivariate omics prediction tasks) and you need to avoid overfitting while maximizing predictive accuracy on held-out test data.

Use when when you have annotated representative LCMS samples (raw mzML files + labeled feature tables in mzmine CSV format) and need to convert them into balanced or unbalanced peak matrix batches with fixed dimensions for neural network training.

Use when when you have a neural network layer definition (parameters, weight initialization, embedding dimension) from a trained or pretrained model checkpoint and need to generate embeddings or activations for a new batch of chemical formulas or spectrum fragments.

Use when after deploying a TensorFlow-backed classification service, you need to verify that the model's input layer names ('input_2048' and 'input_4096') and output layer name ('output') are correctly configured before constructing inference requests.

Use when when you have 512-dimensional (or other fixed-size) representation vectors output from paired encoders processing augmented versions of the same input (e.

Use when when training a deep neural network on paired MS/MS spectra to predict structural similarity scores, especially when the training dataset is moderate-sized (109,734 spectra across 15,062 molecules) and overfitting risk is high.

Use when after training a NeatMS neural network model on labelled peak data (High_quality, Low_quality, Noise) and you need to determine the optimal probability threshold for classifying peaks in your untargeted LCMS dataset.

Use when your input is a corpus of MS/MS spectra with annotated molecular formulas and adduct types that represent a new ionisation mode, instrument type, or adduct chemistry not well-represented in the pre-trained model's training data. You have access to a trained formula transformer (e.

Use when you have trained MLPNN models on paired microbiome-metabolome data (via 10-fold cross-validation repeated across multiple iterations) and need to derive interpretable microbe-metabolite interaction scores rather than treating the network as a black box.

Use when when building Word2Vec or embedding-based spectral similarity models where you need to capture fragmentation patterns beyond individual peak positions.

Use when during MS/MS spectral preprocessing when converting raw spectra from .mgf, .msp, or .mzML formats into a bag-of-fragments corpus for LDA modeling. Use it after fragment ion masses have been extracted and normalized within each spectrum, and before noise filtering and corpus generation.

Use when when converting MS/MS spectra into spectral documents for Spec2Vec embedding, and you want to capture chemical relationships implicit in the fragmentation pattern (e.g., loss of water, ammonia, or CO2) that may correlate with structural similarity. Use this when the neutral loss range (5.

Use when after LDA modeling has inferred a set of Mass2Motifs (in JSON format) from preprocessed MS/MS spectra and you need to annotate these motifs by retrieving matching entries from a MotifDB reference database.

Use when you have a collection of annotated MS/MS spectra with precursor m/z values and fragment peak lists, and you aim to train or apply a spectral similarity model (such as Spec2Vec Word2Vec) that exploits fragmentation chemistry.

Use when you have MS2 spectra data (MGF/mzML format) and aligned feature tables, and your analysis goal is to compare samples that may have poor MS1 feature overlap, strong retention-time shifts across runs, or were acquired on different LC-MS platforms.

Use when you have a tandem mass spectrum (MSMS) loaded via USI and wish to maximize the interpretability of observed peaks.

Use when you have a Chemical Feature Tree artifact (phylogeny) output from q2-qemistree's make-hierarchy method and need to verify its structural validity, count nodes (leaves and internal nodes), measure tree depth, and assess branching patterns before using it for alpha- or beta-diversity.

Use when you have .mzML or .abf LC-HRMS metabolomics raw data files and need to perform peak detection, feature identification, and chromatogram alignment reproducibly across different compute environments (local workstations, HPC clusters).

Use when when you have normalized peak intensity matrices from FT-ICR MS data or other high-resolution metabolomics experiments and need to visualize sample relationships and assess whether categorical grouping factors (e.

Use when you have a trained multitask machine learning model for structure prediction, test set molecules with paired ¹H and ¹³C NMR spectra, and need to understand the marginal contribution of each NMR modality or justify multimodal input design.

Use when when you have raw or lightly processed 1D NMR spectra (¹H and/or ¹³C) from unknown organic compounds and need to extract latent spectral features prior to structure elucidation.

Use when you have 1D NMR spectra (¹H or ¹³C or both) for an unknown organic compound with ≤19 heavy atoms and need to rapidly predict its molecular formula and connectivity graph without manual peak interpretation or exhaustive combinatorial search.

Use when when you have raw or centroid-mode LC-MS All-ion fragmentation (AIF) spectra and need to generate or match against ion fragment databases.

Use when you have metabolomics intensity data with peak annotations, and you want to rank and prioritize metabolite groupings (Molecular Families, Mass2Motifs, or other non-pathway metabolite sets) by their activity levels across experimental contrasts.

Use when you have acquired a mass spectrum from an unknown suspected illicit drug analyte and need to compare it against a synthetic NPS database to rank candidate identities by similarity.