AI & ML

holobiomicslab Skill

Use when after biomolecular class labels have been assigned to features in a TWIM-MS dataset and you have raw ion mobility arrival time measurements. Use it when you need to convert arrival times to standardized CCS values where calibration accuracy depends critically on the biomolecular class (e.

holobiomicslab Skill

Use when after batch correction of a metabolomics dataset using pooled study quality control (SQC) samples, when you have multiple candidate internal standards and need to systematically evaluate which one produces the most stable compound quantification (lowest QC variability) for each compound.

holobiomicslab Skill

Use when after a transformer-based de novo sequencing model (such as Casanovo) generates candidate peptide sequences from MS/MS spectra, before exporting results or using them in database matching or visualization workflows.

holobiomicslab Skill

Use when processing raw MS/MS spectra (in MGF, mzML, mzXML, JSON, or MSP format) prior to MS2Query library matching or MS2Deepscore embedding calculation.

holobiomicslab Skill

Use when you have transformer encoder components and a prediction head specification, and need to wire them into a single trainable model that maps molecular structure inputs (SMILES, molecular graphs, or feature vectors) to scalar or vector molecular property predictions.

holobiomicslab Skill

Use when you have computed a histogram of mass differences from all pairwise mass comparisons in your MALDI-MS imaging dataset and need to prioritize which mass differences are most frequent and likely represent genuine molecular adducts (e.g., metabolite + matrix ions) rather than noise.

holobiomicslab Skill

Use when you have two or more feature tables in HDF5 format with detected features characterized by m/z, drift time, retention time, and intensity, and you need to match corresponding features across samples to account for systematic shifts caused by instrument variation or tuning differences.

holobiomicslab Skill

Use when after executing forward inference on preprocessed mass spectrometry spectra with a deep learning model (e.g., PS²MS), when you have per-spectrum predictions with associated confidence scores or per-class probabilities.

holobiomicslab Skill

Use when you have raw feature tables exported from NPP tools (XCMS, MZmine 2, MS-DIAL, OpenMS, etc.) in their native formats and need to compare their peak detection and alignment performance against a mzRAPP benchmark dataset.

holobiomicslab Skill

Use when you have 512-dimensional representation vectors output from ResNet18 encoders processing paired augmented ion images, and you need to prevent trivial solutions (representation collapse) during contrastive learning—specifically when optimizing for maximized similarity between augmentations.

holobiomicslab Skill

Use when after running formula inference on a benchmark dataset with known formula and adduct ground truth (e.g., NPLIB1, NIST20, or CASMI 2022). Apply this skill when you need to quantify ranking performance, isolate the contribution of specific model features (e.

holobiomicslab Skill

Use when when running metabologenomic RiPP detection pipelines (MetaMiner) on the same genomic dataset but with different input sequence formats (e.g., contigs.fasta vs. antiSMASH .final.gbk output), or when unexpected null results occur and input format choice is a plausible cause.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when after computing Multi-Block Variable Importance in Projection (MB-VIP) scores on a fitted MB-PLS discriminant model, when you need to distinguish signal features from noise by establishing empirical significance thresholds rather than relying on parametric assumptions.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when when training or validating a deep learning model for molecular formula prediction from tandem MS/MS spectra, use this metric to track whether the model's predicted formula (including hydrogen atoms) exactly matches the annotated ground-truth formula.

holobiomicslab Skill

Use when when importing a tab-delimited or Sciex OS text export metabolomics dataset into mzQuality, before building the SummarizedExperiment object.

holobiomicslab Skill

Use when after training or evaluating a classification model (e.g., a Siamese neural network for spectrum similarity prediction) and obtaining a prediction array and corresponding ground-truth label array.

holobiomicslab Skill

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.

holobiomicslab Skill

Use when when you have (1) LC-MS normalized intracellular metabolite abundance data across multiple cell lines or samples, (2) a constraint-based metabolic model with stoichiometric coefficients, and (3) a need to quantify metabolic control through substrate availability independently of enzymatic.

holobiomicslab Skill

Use when use this skill after training or fine-tuning a chemical formula transformer model on annotated tandem MS/MS spectra, when you need to measure whether the model's ranked formula candidates match ground truth.

holobiomicslab Skill

Use when after computing a pairwise sample distance matrix from aligned MS2 fingerprint vectors and you need to visualize sample relationships, clustering, or separation by group identity in 2D space.

holobiomicslab Skill

Use when apply batch normalization after dense hidden layers (but not the final embedding layer) in a deep neural network trained on MS/MS spectral data, particularly when the network processes high-dimensional binned spectra (9948-dimensional vectors) and you need to stabilize gradient flow across.

holobiomicslab Skill

Use when when you have paired predictions and ground-truth structural similarity labels (e.g., predicted Tanimoto scores from a neural network and reference Tanimoto scores from RDKit Daylight fingerprints) and need to report a single scalar metric of model prediction error across all pairs.

holobiomicslab Skill

Use when when you need to compare spectrum prediction models fairly across different encoder architectures (GNN vs. FFN vs. Transformer), and you require equivalent settings (same covariates, identical hyperparameter sweeps) to isolate the effect of the encoder design.

holobiomicslab Skill

Use when when you have a list of SMILES strings representing chemical structures and need to create paired SMILES-spectrum training data for a generative model (like MSGO) without requiring experimental mass spectra.

holobiomicslab Skill

Use when when you have CNN-encoded spectral features (¹H and/or ¹³C NMR) and a set of predicted or candidate molecular fragments, and you need to determine which fragments are present and how they connect to form a valid molecular structure.

holobiomicslab Skill

Use when you have a trained graph neural network model for CCS prediction and need to identify which molecular structural features drive individual predictions or systematic biases.

holobiomicslab Skill

Use when after training a GNN model on molecular structures with continuous targets (e.g., CCS values), when you need to understand which node-level (atom) or edge-level (bond) features contribute most to individual or aggregate predictions.

holobiomicslab Skill

Use when when you have implemented a new or optimized mass spectrometry data processing library and need to demonstrate its computational advantage over established alternatives (e.g., pymzML, pyOpenMS) on real proteomics data. Trigger on availability of: (1) a common input dataset (e.

holobiomicslab Skill

Use when you have preprocessed MSI data (peaks already binned and normalized) and need to detect adduct formation patterns across the dataset.

holobiomicslab Skill

Use when training a contrastive learning model on ion image data (mass spectrometry imaging) where augmented pairs of the same ion image must maximize similarity while different images minimize similarity.

holobiomicslab Skill

Use when when you have a trained multitask model that accepts multiple input modalities (e.g., 1D NMR spectra in different nuclei or complementary analytical techniques) and you need to understand their relative importance for the downstream prediction task (e.g., molecular structure elucidation).

holobiomicslab Skill

Use when a machine learning model produces multiple ranked predictions (each with an associated confidence score) for a single input, and you need to quantify how often the correct answer appears in the top-k predictions.

holobiomicslab Skill

Use when you have a trained formula ranking model (such as MIST-CF) and want to measure the specific performance gain from incorporating multiple positive-mode adduct types (e.g., [M+H]+, [M+Na]+, [M+K]+, [M+NH4]+) instead of restricting predictions to [M+H]+ only.

holobiomicslab Skill

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