Claude Skills by HolobiomicsLab

Use when you have downloaded the LC-MS spectral peak dataset (DOI 10.25345/C5FD2F) and need to train a DNN model from scratch rather than using a pre-trained checkpoint.

Use when you have unpaired mass spectrometry spectra and need to predict Tanimoto-based molecular structural similarity scores between spectrum pairs.

Use when when you have preprocessed MS/MS spectral data (normalized peak intensities and m/z values) and need to convert each spectrum into a learned molecular embedding vector for downstream matching against a reference database.

Use when you have paired tandem MS/MS spectra with known molecular fingerprints, chemical formulae, or SMILES annotations, and you want to learn a generalizable model that can predict molecular properties or annotate unknown spectra by ranking candidates.

Use when you have a pre-trained deep learning encoder (e.g., TCN spectrum encoder trained on a large corpus) and want to adapt it to a new task (e.

Use when training a CNN model from scratch on LCMS peak classification tasks (or similar image-like batched data) where you need to confirm the model reaches target performance (e.g., AUC ROC > 0.9) without overfitting.

Use when when you have a binned MS/MS spectrum vector (e.g., 9948-dimensional input from 10,000 equally-spaced m/z bins in the 10–1000 Da range) and need to compress it into a learned latent representation (e.

Use when before launching the DaDIA pipeline or any multi-package R workflow that has strict version constraints. Use this skill when you have access to an R environment and need to confirm that R ≥4.0, XCMS ≥3.11.4, and metaMS ≥1.25.

Use when before launching a multi-tool computational workflow (e.g., QCxMS2 mass spectra calculations) that depends on external programs with version-sensitive APIs or features. Apply this skill when: (1) the workflow has explicit minimum version requirements for one or more dependencies;

Use when when preparing metabolomics abundance tables with left-censored missingness (values below instrument detection limit or quantification limit) for imputation.

Use when you have raw LC-MS data in mzML format with regions of interest (ROI) already identified, and a trained detection model (e.g., checkpoint0029.pth) has produced bounding box predictions with confidence scores.

Use when you have raw mass spectrometry data from a DIA acquisition method and need to determine whether it is MS^E, All-Ion Fragmentation (AIF), or SWATH-MS before loading into IDSL.CSA for fragmentation spectra deconvolution.

Use when you have raw DIA mass spectrometry files in timsTOF (.d), TripleTOF (.wiff), or Orbitrap (.raw) format and need to extract precursor ion chromatogram (PIC) data as input for quality metric computation or machine learning-based file quality prediction.

Use when you have a batch of DIA mass spectrometry raw files (.raw, .d, or .wiff formats) from known instrument types (timsTOF, TripleTOF, or Orbitrap) and need to classify each file as pass/fail or assign a quality label.

Use when when you have centroided, data-dependent acquisition (ddMS2) spectra in mzML format with extracted precursor m/z values and associated fragment ion lists, and you need to discriminate PFAS features from background signals using characteristic fragmentation patterns.

Use when after running targeted peak detection in TARDIS (screening_mode = FALSE) on centroided .mzML LC–MS files, apply this skill to verify that integrated peaks for your target compounds exhibit acceptable quality.

Use when when analyzing tandem MS/MS spectra with SIMILE V2 and you want to leverage both fragment ion mass differences and neutral loss patterns to improve spectral similarity scoring.

Use when you have normalized and aligned lipidomic and metabolomic spectral features from the Multi-ABLE method across multiple biological samples grouped by phenotype (e.

Use when you have computed high-dimensional embeddings (e.

Use when you have raw or partially organized natural products data from multiple sources (GNPS molecular networking, AntiSMASH BGC predictions, BigScape clustering, MIBiG metadata) and need to prepare them for NPLinker integration.

Use when you have raw LC-MS metabolomics data from multiple disease groups (in .npy or .

Use when when you have detected feature tables from two or more LC-IMS-MS/MS samples and need to establish correspondence between features across samples (i.e., which features in sample A match which features in sample B).

Use when you have a sparse pairwise distance matrix derived from nearest neighbor indexing of MS/MS spectra (or similar high-dimensional objects) and need to partition spectra into groups based on local density and neighborhood connectivity.

Use when you have pre-computed feature-hashed spectrum vectors and nearest neighbor indexes constructed from those vectors, and you need a pairwise distance matrix as input to density-based clustering (e.g., DBSCAN).

Use when after imputation and correction, when metabolomic feature intensities exhibit heteroscedastic variance or non-normal distributions across samples.

Use when apply DModX when you have a normalized LC-MS feature matrix (post-normalization, Step 7 in OUKS) and need to identify samples whose metabolomic profiles are systematically displaced from the learned PCA subspace—indicating potential technical artifacts, extreme biological phenotypes, or.

Use when you have GNPS-style MGF spectral files as input and need to run Mass2SMILES MS/MS-to-structure inference without installing TensorFlow, CUDA, or Python dependencies locally.

Use when when uploading or ingesting paired omics project documents into the Pairing Omics Data Platform, or when you need to verify that a JSON project file conforms to the expected schema structure before processing MS/MS mass spectra linkages, genome associations, or submission to external.

Use when you have acquired one or more tandem MS/MS spectra and need to identify metabolites against a reference library filtered by biological domain (e.

Use when when searching an unknown MS/MS spectrum against a spectral library and you need to rapidly identify the most relevant candidate library spectra before applying cascade scoring with false discovery rate control.

Use when when constructing a comprehensive lipid spectral reference library that must disambiguate lipids differing only in carbon-carbon double-bond position (e.

Use when you have tunemix or other reference standards with known m/z, drift-time, charge state, and CCS values, and you need to establish a predictive calibration model for your ion-mobility mass spectrometry instrument.

Use when you have search results from DIA mass spectrometry data containing feature Q-value scores and need to restrict the analyte choices available to users in a GUI to only those meeting a 1% feature Q-value threshold.

Use when when training a deep neural network on mass spectrometry spectral data where overfitting is a risk (especially with data augmentation applied), and when you need both regularization during training AND uncertainty quantification at inference time via multiple forward passes with dropout.

Use when when you have molecular input data available in two or more distinct formats (e.g., RDKit-extracted fingerprints AND torch_geometric Graph objects representing molecular topology) and your prediction target (e.

Use when you have peak-picked LC-MS data with multiple features that may represent the same metabolite detected at slightly different m/z or retention time values due to instrument precision limits, isotope variants, or adduct forms.

Use when you have a large collection of MS/MS spectra with harmonized metadata and suspect that spectra for the same compound or adduct acquired under similar conditions may be duplicated or near-identical.

Use when you have prototyped a novel data-dependent acquisition strategy that uses dynamic exclusion with intensity or ROI weighting, and you need to quantitatively compare its MS/MS coverage and intensity performance against a simpler baseline (TopN) before testing on real instrumentation.

Use when you have per-sample metabolite abundance data (e.g., from LC-MS or GC-MS) and a metabolite-to-pathway assignment table, and need to generate a sample-by-pathway dysregulation matrix for downstream classification, prognosis prediction, or pathway-level phenotype association.

Use when after running XCMS getEIC() to generate xcmsEIC objects and fillPeaks() to produce a filled xcmsSet object, before computing the 12 peak-quality metrics (Apex Max-Boundary Ratio, Elution Shift, FWHM2Base, Jaggedness, Modality, Symmetry, Sharpness, Gaussian Similarity, Retention-Time.

Use when after completing MS2 annotation in the JPA metabolomics workflow, when you have aligned feature data (feature matrix with m/z, retention time, intensity, and sample assignments) and need to extract and export EIC traces for individual features or feature subsets for external validation.

Use when after abundance-correlation-based feature group refinement when you observe that larger feature groups (particularly those with 3+ features in the same m/z–retention-time window) may contain features with different peak shapes or retention-time shifts in their EICs, or when abundance.

Use when you have raw MS/MS peak lists and suspect electronic noise contamination—particularly when peaks show repeated, identical intensity values across multiple m/z entries within a single spectrum, which are rare in genuine biological spectra but common in instrument artifacts.

Use when when you need to create synthetic noisy MS/MS spectra for benchmarking or validating denoising algorithms.

Use when working with raw MS/MS spectra that contain ions with repeated (identical) intensity values—a hallmark of electronic noise rather than true metabolite fragments.

Use when when a QC check fails during an active LC-MS run and configured email notification targets exist in the system. Use this skill to ensure that QC failures are communicated to stakeholders immediately, complementing Slack-based alerts for users who prefer or require email notification.

Use when when you have pre-computed embeddings (from MSBERT, Spec2Vec, or other deep learning models) for a query spectrum dataset and a reference library, and need to measure how well the embedding space ranks correct library matches.

Use when when you have a collection of MS/MS spectra (in mzML or MGF format) and need to group or retrieve spectra from the same peptide, or when you need to perform large-scale similarity comparisons across millions of spectra without working directly in the original high-dimensional peak space.

Use when when designing a Siamese or multi-branch neural architecture where two or more embedding streams (e.

Use when after training or loading a pre-trained deep learning model that produces high-dimensional embeddings (e.g., 200-dimensional vectors from MS2DeepScore). Use this skill when you have a set of spectra with known chemical class annotations (e.