Claude Skills by HolobiomicsLab

Use when when a new file format specification has multiple language implementations and you need to validate that all implementations correctly interpret the specification.

Use when when you have domain-specific functionality (e.g., spectral similarity scoring, peak detection algorithms) implemented in one language (Python) but need to make it callable and composable within an R-based analytical pipeline (Spectra objects);

Use when you are designing a new tool for FT-ICR MS analysis (or similar high-resolution mass spectrometry domain) and need to understand which analytical and visualization features are already implemented in competing or complementary tools (e.

Use when when cataloging a suite of related bioinformatics tools or web applications (particularly in domains like metabolomics, microbiology, or systems biology) and you need to establish the authoritative peer-reviewed or preprint publication for each tool, verify publication URLs are live, and.

Use when when a deep learning pipeline processes mass spectrometry spectra through multiple independent scripts (e.g., train_rescore.py, run_fiddle.py, test_caffeine.py) and a specific feature must be removed or masked to prevent the model from learning directly from a protected input (e.

Use when when preparing augmented training data for a Siamese rescore model that must learn to rank correct molecular formulas above incorrect ones;

Use when you have executed multiple NPDtools database search pipelines (Dereplicator, VarQuest, Dereplicator+, or MetaMiner in different modes) on identical test spectra or RiPP sequence inputs and need to understand their relative sensitivity, specificity, and complementarity.

Use when you have developed a predictive model and need to compare its performance against established baselines (e.g., linear regression, Random Forest, Canonical Correlation Analysis) across multiple datasets with paired input-output features.

Use when when you have downloaded a multi-file .csv library repository (e.g., LipidMatch) and need to verify that it meets minimum thresholds for species diversity (e.g., 500,000+ distinct lipid species) and category breadth (e.g., 60+ lipid-type categories).

Use when when standard conversion directives (headers, collate, fields_to_headers, exclusion_headers, values_to_str, sort_by, test) cannot express the required transformation logic, or when domain-specific aggregation, conditional logic, or value derivation must be applied to records after field.

Use when a practitioner has pre-computed features from an external feature-finding procedure (e.g., vendor software, alternative open-source tools) and wishes to incorporate them into PFΔScreen's PFAS prioritization pipeline without re-detecting features from raw mzML data.

Use when when you have mass spectrometry data stored in a database or non-mzML file format and need to integrate it into the pymzML pipeline without converting to mzML. Use this skill when FileInterface does not natively recognize your file extension and you want to avoid data conversion overhead.

Use when you have a user-supplied metabolite set file (CSV or JSON) defining custom groupings of metabolites (e.

Use when after computing pairwise mass-difference transformations between FT-ICR MS peaks and matching them to a reference biochemical transformation key, you have putative edge data (source peak, target peak, transformation type, mass error) and node data (peaks with m/z, molecular formula.

Use when after loading a specXplore session data object (saved .

Use when you have a conceptual MS/MS fragmentation strategy (e.

Use when you have retrieved a complete set of project JSON documents from a data platform and need to verify that each document's structure, field types, and required properties match a canonical JSON Schema definition (e.g., app/public/schema.json).

Use when you have Spectra::Spectra objects in R and need to apply Python MS algorithms from matchms or spectrum_utils (e.g., CosineGreedy similarity scoring, normalization, or filtering), or conversely need to convert Python spectral objects back to R for downstream Spectra workflows.

Use when you have a new or draft file format specification (e.g., mzPeak) with multiple independent language implementations, and you need to verify that all readers agree on the structured data they extract before recommending the format for production use.

Use when after obtaining structural clusters from the MAMSI framework using different parameter configurations (e.

Use when you have a curated relational dataset (structure-organism pairs) and need to quantify how structures distribute across a categorical variable (e.g., organism prevalence).

Use when you have raw microbiome (e.g., 16S rRNA or metagenomic) or metabolomic count tables (samples × features) and plan to train predictive models (e.g., MiMeNet, linear regression) that require normally distributed or near-normally distributed inputs.

Use when building or auditing a multi-instrument MS data processing system that must route different chromatography modes (LC, GC), ion mobility, or imaging modalities (MALDI) to distinct processing workflows.

Use when a Python module declares optional/conditional dependencies (e.g., sqlalchemy for database access) and you need to confirm that the module can be imported and instantiated without exceptions when those dependencies are present in the environment.

Use when when you have downloaded a curated structure-organism dataset (such as LOTUS) and need to verify the reported counts of unique entities (source databases, organisms, structures, and their pairs) to confirm dataset integrity, assess data coverage, or reproduce published statistics in a.

Use when when you have an mzML file that you want to store persistently in a queryable format for repeated access, or when memory constraints prevent loading entire mzML files into memory;

Use when when you have large sequential media files (mzML spectra, text chapters, or similar) and need to enable both random access by ID and sequential iteration without loading the entire file into memory.

Use when you are implementing a new MsBackend subclass and need to store spectra metadata (sample names, retention times, precursor m/z, etc.) separately from peak data (m/z and intensity pairs) while maintaining row-wise alignment.

Use when when implementing a new MsBackend subclass that stores only a subset of core spectra variables (e.

Use when when designing or optimizing an MsBackend implementation (or similar columnar data structure) you must decide whether to pre-allocate all known columns in the backing DataFrame at initialization or defer column creation until first access.

Use when when building a scientific visualization library that must support multiple plotting backends and needs to avoid backend-specific code duplication. Specifically: (1) your domain (e.

Use when when you have downloaded a released version of a structured dataset (e.g., LOTUS from Zenodo) and need to confirm it matches the documented headline statistics before downstream analysis, or when auditing data integrity after ingestion into a processing pipeline.

Use when when you have a raw GNPS or other spectral library dataset with inconsistent or incomplete instrument annotations, and you need to verify or reproduce reported dataset split counts (e.g., training/test compound ratios). Apply this skill when an instrument allowlist fix (e.

Use when you have a virtual chemical mixture (MS1 peaks) and need to prototype a new DDA acquisition strategy before testing on real instrumentation. Use this skill when you want to compare how different parameter combinations (e.g., TopN=3 vs TopN=5, isolation_width=0.5 Da vs 1.

Use when you have MS/MS spectra (centroided m/z and intensity pairs) and corresponding MS1 precursor masses but lack reference spectra or a priori formula information.

Use when when you have unknown MS/MS spectra with observed precursor m/z values and want to infer the molecular formula and adduct type (e.g., [M+H]+, [M+Na]+, [M+K]+) in a de novo setting without access to spectral libraries.

Use when when you have high-resolution LC-MS/MS data for an unknown metabolite or small molecule, have computed or measured the molecular ion mass and fragmentation spectrum, and require de-novo structure generation because the compound is absent from spectral libraries or structure databases.

Use when you have two augmented versions of the same ion image (from mass spectrometry imaging data) and need to extract learnable 512-dimensional feature representations using a shared-weight encoder for contrastive loss optimization.

Use when after converting or loading a Keras model to HDF5 TensorFlow 2.3.0 format, especially when the model will be served through a pipeline (e.g., NP Classifier) that expects specific input/output layer names.

Use when when you have MS/MS spectra from GNPS or other libraries and need to apply a pre-trained FIDDLE model (TCN formula predictor or Siamese rescore architecture) without training from scratch. Use this skill before running inference on new samples or benchmarks.

Use when you have a trained deep learning model and want to quantify prediction uncertainty for each input pair or decision point. Use this when you need to identify low-confidence predictions (high IQR) and filter them out to reduce error in specific score ranges (e.

Use when you have a pretrained deep learning model, a reserved test set with ground-truth annotations, and need to evaluate prediction quality or generate embeddings for downstream analysis. Typical triggers: benchmarking a new model against classical baselines (e.

Use when you have preprocessed mass spectrometry spectra (tokenized m/z and intensity pairs or feature matrices) and a trained deep learning model checkpoint, and you need to classify unknown compounds or generate prediction confidence scores for structural novelty analysis.

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 you have cloned or loaded a deep-learning architecture extension (e.g., chemprop-IR) and need to verify that its feature extraction component can be instantiated and produces correctly shaped feature tensors before integrating it into a larger pipeline or training loop.

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 you have a pre-trained DNN model for retention time prediction and need to adapt it to a new chromatographic method or instrument where you have only 10–20 calibration molecules with known retention times;

Use when training a DNN on retention time prediction or similar continuous regression tasks where: (1) the feature space is very high-dimensional (thousands of molecular descriptors and fingerprints), (2) the training set is moderately sized (tens of thousands of molecules), (3) you observe or.

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 you have a derivatizing matrix (e.g., TAHS or other publicly documented reagent) with known composition and ionization behavior that you want to use in Met-ID for metabolite annotation, and the matrix is not yet configured in your Met-ID installation.