AI & ML

holobiomicslab Skill

Use when when you have a neural network or machine learning model with multiple tunable hyperparameters (layer size, regularization strength, dropout) or design choices (e.

holobiomicslab Skill

Use when you have a GC-MS dataset with multiple sample files (e.g., Std_soln_00, Std_soln_07, Std_soln_00a) where the same chemical is detected in different runs with varying Match.Factor scores, and you need to consolidate area values by compound identity rather than by individual peak detection.

holobiomicslab Skill

Use when you have curated a labeled dataset of MS/MS spectra annotated as 'relevant' (compounds of interest obtained from reference standards) or 'other' (reference standards or non-target compounds), and you need to build a classifier to automatically distinguish these two classes on unknown.

holobiomicslab Skill

Use when when you need to understand how a multi-instrument mass spectrometry platform (like mzmine) decides which processing module receives a given dataset based on its declared data type (LC vs. GC vs. IMS vs. MS imaging).

holobiomicslab Skill

Use when you have a generic constraint-based metabolic model (SBML format) and cross-sectional omics data (RNA-seq, intracellular metabolomics, YSI or bioanalyzer extracellular flux measurements) for multiple biological samples and need to create sample-specific models that discriminate whether.

holobiomicslab Skill

Use when you have raw SMILES strings from a chemical database (e.g., CCSBase, METLIN, or custom sources) and need to feed them into a graph neural network model.

holobiomicslab Skill

Use when when you have a training dataset of NMR spectra-derived molecular fragments encoded as token sequences, and you need to predict both the molecular formula and complete bond connectivity of unknown molecules.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when you have a collection of preprocessed tandem mass spectra (binned into 10,000 equally-sized m/z bins, intensities square-root transformed, top 1,000 peaks retained), a trained MS2DeepScore Siamese model, and you need to predict structural similarity scores (Tanimoto or Dice) for all or a.

holobiomicslab Skill

Use when you have implemented or modified a tandem mass spectrometry formula inference model and need to measure whether a specific architectural change (e.

holobiomicslab Skill

Use when when training a formula-prediction model with a frozen pretrained TCN spectrum encoder and unfrozen FormulaEncoder and RescoreHead components, or when you suspect the model may use precursor intensity as a spurious feature rather than fragment-pattern information for molecular formula.

holobiomicslab Skill

Use when you have tandem mass spectra (in .msp or compatible format) from instruments like Orbitrap, and you need to compute library matching scores, cluster spectra by chemical similarity, or embed spectra into a learned vector space for downstream similarity or clustering tasks.

holobiomicslab Skill

Use when you have a trained GNN model (stored as .h5 weights) and molecular graph representations (SMILES strings and/or 3D coordinates), and you need to compute predicted CCS values or perform feature importance analysis via ablation or gradient-based saliency mapping.

holobiomicslab Skill

Use when you have: (1) a trained IOKR model mapping from spectrum kernels to molecular fingerprints, (2) MS2 spectra from your sample, (3) a set of candidate BGCs with known or predicted structures (e.

holobiomicslab Skill

Use when you have developed a new machine learning model for predicting metabolomic profiles from microbiome data and need to quantify its performance improvement over existing methods.

holobiomicslab Skill

Use when you have measured intracellular metabolite abundances (LC-MS or similar) across multiple cell lines or conditions and a stoichiometric metabolic model (with reaction-metabolite associations) to estimate how differences in substrate availability—independent of gene expression—translate into.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when you have a pretrained TCN spectrum encoder from formula prediction and need to train a rescoring model that ranks formula candidates by confidence. The input is a set of spectra with ground-truth formula labels and multiple candidate formulas per spectrum.

holobiomicslab Skill

Use when after training a GNN-RT model on preprocessed molecular graph data (from Train.py) or after applying transfer learning to an in-house dataset (from Transferlearning.

holobiomicslab Skill

Use when you have trained a multitask NMR-to-structure model and need to quantify its predictive accuracy on held-out test molecules.

holobiomicslab Skill

Use when you need to determine the complete set of validated instrument/vendor and acquisition mode combinations for a mass spectrometry analysis tool, when assessing whether your specific instrument platform (vendor, model, acquisition method) is supported before committing to a workflow, or when.

holobiomicslab Skill

Use when you have a pre-trained or newly retrained graph neural network for collision cross section prediction and need to measure whether its performance generalizes across different molecular datasets (e.g., training on METLIN but evaluating on CCSBase).

holobiomicslab Skill

Use when you have (1) a benchmark dataset of reference peaks with validated m/z, retention time boundaries, and isotopologue assignments, and (2) NPP output feature tables (unaligned and aligned) from tools like XCMS, MZmine 2, or MS-DIAL that you wish to evaluate.

holobiomicslab Skill

Use when you have tabulated pairwise mass differences from a MALDI-MS imaging dataset (via massdiff()) and need to identify which observed mass differences correspond to known chemical adducts (e.g., [M+H]+, [M+Na]+, [M-H2O]+).

holobiomicslab Skill

Use when after a machine learning model has generated predicted molecular structures (connectivity graphs and molecular formulas) from 1D NMR spectra. Use it to quantify accuracy on a held-out test set, measure degradation when applying the model beyond its training scope (e.

holobiomicslab Skill

Use when when you have a trained or untrained chemprop base model (graph convolution + readout layers) and need to extend it to predict infrared spectral properties rather than scalar molecular properties.

holobiomicslab Skill

Use when you have a trained spectral embedding model (e.g., MSBERT, Spec2Vec) and need to benchmark its library matching accuracy against reference spectra on a test dataset.

holobiomicslab Skill

Use when when you have 1D NMR spectra (1H and/or 13C) of an unknown compound with up to ~19 heavy atoms and need to predict both molecular formula and connectivity without manual structure hypothesis generation.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when you have paired MS/MS spectra from unknown and known metabolites with raw intensity values and need to prepare them as input for a deep-learning model that will predict structural similarity.

holobiomicslab Skill

Use when you have a generic genome-scale metabolic model (SBML format) and cross-sectional omics data (RNA-seq, intracellular metabolomics, extracellular flux measurements from bioanalyzer or similar) from multiple biological samples (cell lines, conditions).

holobiomicslab Skill

Use when after computing all pairwise mass differences from a mass spectrometry imaging dataset, when you need to identify which mass differences correspond to real molecular adducts (e.g., metabolite–matrix or metabolite–salt ions) rather than noise.

holobiomicslab Skill

Use when you have preprocessed MS/MS spectra pairs (unknown and known metabolites) with annotated structural similarity labels, and you need to learn a generalizable model that can rank candidate structures for novel unknowns by predicting their similarity to reference compounds in a database.

holobiomicslab Skill

Use when when you have a pre-trained Casanovo model, annotated MS/MS spectra in MGF format, and want to benchmark whether beam search decoding improves peptide prediction quality over the default greedy decoding strategy.

holobiomicslab Skill

Use when you have trained two or more graph neural network models on the same CCS dataset split (using identical hyperparameters, loss functions, and optimization settings) and need to rigorously compare their held-out test performance to determine which architecture balances prediction accuracy.

holobiomicslab Skill

Use when you have tandem mass spectra (MS/MS) with unknown precursor formulas and need to rank chemical formula candidates conditioned on observed fragment m/z values and precursor mass. Use this skill when fragmentation tree computation (e.