Claude Skills by HolobiomicsLab

Use when when you have LC-MS data in Thermo .raw format (a proprietary binary output from Thermo mass spectrometers) and need to load it into a 3D LC-MS viewer or extract structured spectral metadata (retention time, m/z values, intensities, scan number, MS level) for visualization or quantitative.

Use when when deploying Mass2SMILES inference on CPU-only hardware (e.g., when GPU/CUDA support is unavailable), you need to verify that the InferenceModel cpu_threads parameter (e.g., cpu_threads=128) actually controls thread allocation during execution.

Use when you have downloaded or cloned a fragmentation library repository (such as LipidMatch) and need to verify that it contains the expected breadth of coverage across both molecular diversity (distinct species count) and chemical classification (lipid-type category count).

Use when after normalization (step 7) in untargeted metabolomic profiling pipelines, when you have a preprocessed feature matrix and need to identify samples with anomalous metabolic profiles that violate multivariate assumptions or represent technical failures.

Use when you have experimental MS/MS spectra that must be matched against a large hierarchical fragmentation library (e.g., 168.

Use when after feature detection when you have a metabolomics feature table with intensity values across multiple samples and observe evidence of inter-sample intensity bias (e.g., batch effects, variable ionization efficiency, or instrument drift).

Use when when deploying a metabolomics processing tool (such as asari) and needing to predict resource requirements or validate claimed scalability on laptop-class hardware (≤16 GB RAM, single CPU core).

Use when you have time-resolved DBDI-MS data (intensity matrix with m/z features as rows and scan timepoints as columns) from direct injection analysis where chromatographic separation is unavailable and you need to identify which features represent fragments or adducts of the same parent compound.

Use when when LCMS metabolomics abundance tables show systematic intensity drift across the injection sequence (e.g., instrument signal decay or gain over hours), and you have pooled technical replicate injections and/or internal standard compounds distributed throughout the run.

Use when you have LC-IM-MS/MS raw data from multiple tissue samples and need to identify and quantify unsaturated sterol lipids at the isomer level (distinguishing double-bond position and stereochemistry).

Use when your raw TOF-MS data (Agilent MassHunter .d format) exhibits jagged, artifact-prone peaks in low-abundance ions that compromise peak quality assessment or when you need to improve signal-to-noise before ion mobility demultiplexing or peak deconvolution.

Use when when you have deployed a trained embedding or similarity model on a test set of tandem mass spectra and need to measure its ability to rank correct library compounds near the top of retrieved candidates.

Use when you have a ranked candidate list (e.g., BGCs sorted by IOKR or strain-correlation score) for each test spectrum, a known ground-truth BGC for each spectrum, and you want to measure retrieval performance across multiple recall depths (top-1 through top-200) and overall discrimination.

Use when when preparing to apply LDA to a new MS/MS spectral dataset in bag-of-fragments format, before running the full modeling pipeline.

Use when after generating a corpus/features JSON file from raw MS2 fragmentation spectra or similar document-term data, when the goal is unsupervised discovery of latent topics (e.g., metabolite families or spectral motifs) without ground-truth labels.

Use when when you have real LC-MS/MS data (mzML format) from an untargeted metabolomics experiment and want to test how variations in TopN DDA parameters affect which precursor ions are selected and fragmented, before deploying the optimized strategy on physical instruments.

Use when after loading all MS1 scans from a raw or intermediate mass spectrum file (e.

Use when when processing LC-MS data from isotope labeling experiments where the tracer (13C, 2H, 15N, 18O, or 34S) has known isotopic impurity and you observe discrepancies between measured isotopologue abundances (FAM) and expected labeling patterns.

Use when you have acquired LC-MS peak tables from both unlabeled (12C) and isotope-labeled (13C) samples from a stable isotope tracing experiment, paired with sample metadata distinguishing case and control groups, and you need to systematically identify which unlabeled features correspond to.

Use when after applying a configuration fix (e.g., adding an instrument type to an allowlist, updating filtering thresholds) to a dataset preprocessing pipeline, you need to confirm that the change produces the documented training/test split counts.

Use when you are preparing to apply Probability Product Kernel–based scoring to MS2 spectra for genomic–metabolomic linking, and you need to establish a reference set of ion peaks that represent robust spectral diversity without overfitting to any single study.

Use when you have downloaded raw LC-MS spectral peak data from a public repository (e.g., DOI 10.25345/C5FD2F) and need to ingest it into memory and prepare it in the format expected by a TensorFlow/Keras neural network classifier.

Use when when you have a complete labelled MS/MS spectral dataset annotated as 'relevant' (compounds of interest from reference standards) or 'other' (reference standards or non-target compounds from repositories), and you need to train and evaluate an AnnoMe binary classifier without data leakage.

Use when after labeling a representative subset of peaks (typically 10–20 pooled samples with corresponding feature tables) and before neural network training, when you need to split labeled data into independent subsets for model training, hyperparameter tuning, and unbiased performance evaluation.

Use when you have a pretrained spectrum encoder (e.g., TCN on mass spectrometry data) that has learned useful representations, and you need to train new components (e.g., a rescoring module) for a related but distinct task (e.

Use when you have collected liquid chromatography (LC) spectra and retention time labels for your in-house molecular database, and you want to leverage a pretrained GNN-RT model rather than train from scratch.

Use when when you have paired microbiome (16S rRNA or functional) and metabolome (LC-MS/MS or similar) data and must decide between compositional transformations (CLR, RA, or others) before training a predictive model.

Use when after generating or obtaining a TP object (via generateTPs with library, in-silico, or BioTransformer algorithms) that contains SMILES, InChIKey, or formula information for transformation products, and you need to screen those TPs in LC-HRMS data or annotate them using MetFrag.

Use when after suspect screening has identified both parent features (from before-treatment or reference samples) and TP candidate features (from after-treatment or exposed samples) in the same analysis set, and you have MS/MS spectral data or formula annotations available.

Use when after parent chemical suspects have been identified in a non-target screening workflow, use this skill when you need to screen for downstream products formed by chemical or biological transformation.

Use when when building a neural network to map between mass spectrometry spectra and molecular properties (e.g., fingerprints, SMILES, or fragment ions) where sequential or spectral feature dependencies must be captured.

Use when when you need to learn chemical-rational embeddings of tandem MS/MS spectra for library matching or molecular property prediction, and you want to leverage self-supervised learning through masking.

Use when you have paired tandem MS spectra and either (1) molecular fingerprints or structures as labels for supervised fingerprint prediction, or (2) both spectra and unpaired structure/SMILES libraries and want to train embeddings for database-free structure lookup.

Use when when you have preprocessed MS/MS spectra (normalized intensities, filtered for quality, with top peaks retained) that have been encoded using a spectral representation method (e.

Use when preparing chemical formulae (e.g., 'C6H12O6') as inputs to a transformer-based neural network for MS/MS spectrum scoring. Use it specifically when the transformer must rank multiple candidate formulae against an observed mass spectrum and element composition constraints are important;

Use when you have variable-length MS/MS peak lists (m/z arrays and intensity arrays) that must be fed into a transformer architecture for tasks like compound identification or spectral clustering.

Use when you have a dataset of augmented simulated overlapped GC-MS peaks and need to train a Transformer model to automatically deconvolve them into pure component mass spectra.

Use when when processing raw FT-ICR transient files (Bruker Solarix .d format or equivalent) intended for high-resolution mass spectral analysis.

Use when when you have loaded a TransitionGroup (extracted ion chromatogram or mobilogram from DIA mass spectrometry data) and need to identify precise peak boundaries, apex retention/drift time, and intensity values for quantitative feature detection.

Use when you have loaded extracted ion chromatogram (XIC) data from DIA mass spectrometry and need to identify peak boundaries for peptide precursor transitions.

Use when you have generated or assembled a lipid spectral library with precursor m/z values, adduct information, and fragmentation patterns, and you need to import those spectra into Skyline for targeted data-independent or parallel-reaction-monitoring (PRM) analysis.

Use when after training a decision tree classifier on ChemEcho sparse feature vectors (representing tandem mass spectra fragmentation patterns), especially when the goal is to deploy predictions as executable queries or to validate that learned splits correspond to chemically meaningful.

Use when you have a connected subnetwork of LC-MS features that matched isotope or adduct patterns, and you need to establish a canonical tree representation with a single neutral mass assignment.

Use when you have a shallow decision tree trained on ChemEcho feature vectors (peak or neutral loss formulas) and need to deploy it as an executable query against tandem mass spectrometry data.

Use when after applying one or more mpactr filters (filter_mispicked_ions, filter_group, filter_cv, filter_insource_ions) to an mpactr object, use this skill when you need to communicate the count and percentage breakdown of ions retained vs. rejected across filter status categories.

Use when you have a trained NeatMS neural network model (.h5 format) and need to assess its classification performance at a specific decision threshold (e.g., 0.01) to determine what fraction of true peaks are retained (TPR) versus what fraction of incorrect peaks are incorrectly accepted (FPR).

Use when when you have a precomputed similarity matrix of mass spectra (e.

Use when you have raw GCxGC-MS chromatogram data in NetCDF format from multiple samples (e.g., case and control groups) and need to prepare them for multivariate analysis such as multiway principal component analysis (MPCA).

Use when you have raw GC–MS or LC–MS data represented as a two-dimensional map (m/z axis vs. retention time axis) and need to identify chemo-/biomarker features across multiple analytes simultaneously, especially when conventional peak picking produces high false-positive or false-negative rates.

Use when you have GC–MS or LC–MS data represented as a two-dimensional map with m/z values on one axis and retention time on the other, and you need to identify analyte signals and chemo-/biomarker features while minimizing false positive and false negative peak detections.