Awesome-SpectraAI-Resources

✨✨ A curated collection of resources on artificial intelligence for spectral data analysis, covering computational methods for mass spectrometry (MS), NMR, IR, and XRD data.

---

• 1. Mass Spectrometry (Small Molecules)
• 2. Mass Spectrometry (Peptides)
• 3. NMR Spectroscopy (Small Molecules)
• 4. IR Spectroscopy (Small Molecules)
• 5. Multimodal Spectroscopy (Small Molecules)
• 6. X-ray Diffraction (XRD) (Crystals)

1. Mass Spectrometry (Small Molecules)

1.1 Forward Task (Molecule → Spectrum)

Computational approaches for predicting mass spectra from molecular structures

Paper Title & Link	Method Type	Venue	Code	Notes
Inferring CID by Estimating Breakage Events and Reconstructing their Graphs	GNN, Transformer	Analytical Chemistry, NeurIPS2023		ICEBERG, SCARF
FIORA: Local neighborhood-based prediction of compound mass spectra from single fragmentation events	GNN	Nature Communication		FIORA
Efficiently predicting high resolution mass spectra with graph neural networks	GNN	ICML2023		GRAFF-MS
CFM-ID 4.0: More Accurate ESI MS/MS Spectral Prediction and Compound Identification	Machine Learning Model	analytical chemistry		CFM-ID 4.0, project website
Computational prediction of electron ionization mass spectra to assist in GC-MS compound identification	Artificial Neural Network, Probabilistic Generative Model	analytical chemistry		CFM-EI,old version of CFM-ID, project website
Tandem mass spectrum prediction for small molecules using graph transformers	Graph Transformer, Deep Learning Model	Nature Machine Intelligence		MassFormer
Rapid Prediction of Electron-Ionization Mass Spectrometry Using Neural Networks	Neural Network, Graph-Convolutional Network	ACS Central Science		NEIMS
Mass Spectra Prediction with Structural Motif-based Graph Neural Networks	GNN, MLP, Graph Transformer	Scientific Reports		MoMS-Net
Prediction of electron ionization mass spectra based on graph convolutional networks	GCN, MLP	International Journal of Mass Spectrometry		Baojie Zhang mass spectra prediction
QCxMS and QCEIMS related publications	Born-Oppenheimer Molecular Dynamics			QCxMS, QCxMS2, QCEIMS, First Principles Calculation, project website
3DMolMS: prediction of tandem mass spectra from 3D molecular conformations	3D Molecular Network	Bioinformatics		3DMolMs,project website
Rapid Approximate Subset-Based Spectra Prediction for Electron Ionization–Mass Spectrometry	GNN	Analytical Chemistry		rassp
Towards First Principles Calculation of Electron Impact Mass Spectra of Molecules	Mass Spectroscopy computational therory	Angewandte International Edition Chemie		theory introduction

1.2 Inverse Task (Spectrum → Molecule)

AI methods for molecular identification and elucidation from mass spectra

Paper Title & Link	Method Type	Venue	Code	Notes
An end-to-end deep learning framework for translating mass spectra to de-novo molecules	GRU, CNN	communications chemistry		Spec2mol
Searching molecular structure databases with tandem mass spectra using CSI:FingerID	Machine Learning	Proceedings of the National Academy of Sciences	project website	CSI:FingerID
Deep kernel learning improves molecular fingerprint prediction from tandem mass spectra	DNN, Kernel-based Model, SVM	Bioinformatics	project website	Deep kernel learning method
CSU-MS $^2$ : A Contrastive Learning Framework for Cross-Modal Compound Identification from MS/MS Spectra to Molecular Structures	Contrastive Learning Framework, Transformer, GNN	Analytical Chemistry		CSU-MS²
MassGenie: A Transformer-Based Deep Learning Method for Identifying Small Molecules from Their Mass Spectra	Transformer, Variational Autoencoder	Biomolecules		MassGenie, FragGenie, VAE-Sim
JESTR: Joint Embedding Space Technique for Ranking Candidate Molecules for the Annotation of Untargeted Metabolomics Data	Joint Embedding Space Technique, GNN, MLP	Bioinformatics		JESTR
Metabolite Identification through Machine Learning — Tackling CASMI Challenge Using FingerID	Machine Learning, Kernel-based approach, SVM	Metabolites		FingerID
Annotating metabolite mass spectra with domain-inspired chemical formula transformers	Transformer, Neural Network, Contrastive Learning Model	Nature Machine Intelligence		MIST
Self-supervised learning of molecular representations from millions of tandem mass spectra using DreaMS	Transformer, Foundation Model	Nature Biotechnology		DreaMS, Foundation model
Systematic classification of unknown metabolites using high-resolution fragmentation mass spectra	DNN, SVMs	Nature Biotechnology		CANOPUS
Automatic Compound Annotation from Mass Spectrometry Data Using MAGMa	Software, Substructure-based algorithm	Mass Spectrometry	project website	MAGMa
DiffMS: Diffusion Generation of Molecules Conditioned on Mass Spectra	Transformer, Discrete Graph Diffusion Model	arXiv		DiffMS
MetFID: artificial neural network-based compound fingerprint prediction for metabolite annotation	Artificial Neural Network	Metabolomics		MetFID
MADGEN - MASS-SPEC ATTENDS TO DE NOVO MOLECULAR GENERATION	Attention-based generative model, LSTM, CNN	arxiv		MADGEN
An end-to-end deep learning framework for translating mass spectra to de-novo molecules	Deep Learning architecture, Autoencoder	Communications Chemistry		Spec2Mol
DiffSpectra: Molecular Structure Elucidation from Spectra using Diffusion Models	Diffusion Model, Transformer	arXiv		DiffSpectra
MS2Mol: A transformer model for illuminating dark chemical space from mass spectra	Transformer, Generative Model			MS2Mol
Improved metabolite identification with MIDAS and MAGMa through MS/MS spectral dataset-driven parameter optimization	Machine Learning, Rule-based	Metabolomics		MIDAS, MAGMa, CSI: FingerID
Using Graph Neural Networks for Mass Spectrometry Prediction	GNN	arXiv		GNN-based models
MSNovelist: de novo structure generation from mass spectra	Encoder-decoder neural network, RNN	Nature Methods		MSNovelist
Predicting a Molecular Fingerprint from an Electron Ionization Mass Spectrum with Deep Neural Networks	Deep Neural Network	analytical chemistry		DeepEI, FP Model, molecular fingerprint prediction
Deep MS/MS-Aided Structural-Similarity Scoring for Unknown Metabolite Identification	Deep Neural Network			DeepMASS, mass spectrum-based library match for Metabolite
Ultra-fast and accurate electron ionization mass spectrum matching for compound identification with million-scale in-silico library	Word2vec, HNSW	nature communications		FastEI, spectrum simulation expansion, spectrum search for compoundidentification
In silico fragmentation for computer assisted identification of metabolite mass spectra	Combinatorial Fragmenter	BMC Bioinformatics		MetFrag, mass spectrum-based metabolite identification, Project Link
MS2Query: reliable and scalable MS2 mass spectra-based analogue search	embedding-based chemical similarity predictors	Nature Communications		MS2Query, mass spectra analogue search, Project Link: https://doi.org/10.5281/zenodo.6124553

1.3 General Tools

Paper Title & Link	Feasible scene	Venue	Code	Notes
matchms- processing and similarity evaluation of mass spectrometry data	raw mass spectra to pre- and post-processe	The Journal of Open Source Software		python package
MS2DeepScore: a novel deep learning similarity measure to compare tandem mass spectra	compare tandem mass spectra	Journal of Cheminformatics
Spec2Vec: Improved mass spectral similarity scoring through learning of structural relationships	NLP-inspired Model, Word2Vec	PLOS Computational Biology		Spec2Vec
Chemically informed analyses of metabolomics mass spectrometry data with Qemistree	Machine Learning, Tree-based approach	Nature chemical biology		Qemistree
MetaboAnalystR 4.0: a unified LC-MS workflow for global metabolomics	Software Tool	Nature Communications		MetaboAnalystR 4.0, LC-MS data processing, R package, Project Link: MetaboAnalyst
Fully Automated Unconstrained Analysis of High-Resolution Mass Spectrometry Data with Machine Learning	Decision Trees, Neural Network, LSTM	Journal of the American Chemical Society		MEDUSA, mass spectrum analysis overall framework, Project Link: project_link
The METLIN small molecule dataset for machine learning-based retention time prediction	Deep Learning	Nature Communications		project website, retention time prediction, R package
An end-to-end deep learning method for mass spectrometry data analysis to reveal disease-specific metabolic profiles	Deep Learning	Nature Communications		DeepMSProfiler, disease-specific metabolic profiling
Trackable and scalable LC-MS metabolomics data processing using asari	open-source software tool	Nature Communications		asari, LC-MS data processing, Project Link: https://pypi.org/project/asari-metabolomics/
Automatic Compound Annotation from Mass Spectrometry Data Using MAGMa	Ranking algorithm	Mass Spectrom (Tokyo)		Peak list annotation, MAGMa
MIST-CF: Chemical Formula Inference from Tandem Mass Spectra	Spectrum Transformers	Journal of Chemical Information and Modeling		Mass Spectra Formula Inference
Systematic classification of unknown metabolites using high-resolution fragmentation mass spectra	Deep Neural Network	Nature Biotechnology		compound class annotation, CANOPUS

1.4 Datasets, Benchmark and Review

Name	Type	Venue	Website	Notes
MassSpecGym: A benchmark for the discovery and identification of molecules	Benchmark, Transformer, GNN	Advances in Neural Information Processing Systems		MassSpecGym
Sharing and community curation of mass spectrometry data with Global Natural Products Social Molecular Networking	Mass Spectrum Database	Nature Biotechnology		Database link
A versatile toolkit for drug metabolism studies with GNPS2: from drug development to clinical monitoring	Mass Spectrum Database	Nature Protocals		Database link
MassBank: a public repository for sharing mass spectral data for life sciences	Mass Spectrum Database	Journal of mass spectrometry		Database link-EU Database link-NA
Artificial Intelligence in Spectroscopy: Advancing Chemistry from Prediction to Generation and Beyond	Neural architectures, ML-empowered solution	arXiv		review
BMDMS-NP: A comprehensive ESI-MS/MS spectral library of natural compounds	Spectral Library	Phytochemistry		BMDMS-NP, ESI-MS/MS spectral library
Annual Review of Analytical Chemistry Machine Learning in Small-Molecule Mass Spectrometry	Sequence-based Model, Graph-based Model, Deep Learning Model, Siamese Neural Network, Natural Language Processing Model, Transformer, MLP, Random Forest, Bayesian Regularized Neural Network, XGBoost, Light Gradient Boosting Machine, CNN, SVR	Annual Review of Analytical Chemistry		Review Paper, SELFIES, Graph Neural Networks, MPNNs, SchNet, DimeNet++, ComENet, DeepMASS, MS2DeepScore, Spec2Vec, CLERMS, NEIMS, MassFormer, 3DMolMS, CFM-ID 4.0, SCARF, ICEBERG, Retip, METLIN-DLM, GNN-RT, DeepGCN-RT, RT-transformer, DNNpwa-TL, MetCCS, AllCCS, CCSBase, CCSP 2.0, DeepCCS, SigmaCCS, AllCCS2, SIRIUS, BUDDY, mass spectrometry analysis
Critical Assessment of Small Molecule Identification	Competation, Benchmark	Phytochemistry Letters		CASMI 2012, 2013, 2014, 2016, 2017, 2022, project website
Insights into predicting small molecule retention times in liquid chromatography using deep learning	SVM, Deep Learning, Transformer, GNN, RF, MLP, DNN, CNN, RNN	Journal of Cheminformatics		CSI-FingerID, SIRIUS 4, MSNovelist, MassGenie, Smiles-Bert, Smiles transformer, Chemformer, PredRet, GCN, RGCN, MPNN, GIN, DNNpwa-TL, CMM-RT, 1D CNN-TL, AWD-LSTM, TransformerXL, MDC-ANN, retention time_GNN, QGeGNN, RT-transformer, mt-QSRR, MultiConditionRT, HighResNPS, GNN-RT-TL, Retip, SMRT, GNN-TL, review paper
Quantum chemical electron impact mass spectrum prediction for de novo structure elucidation: Assessment against experimental reference data and comparison to competitive fragmentation modeling	Quantum Chemical Model, Expert system, Spectrum Calculator	International Journal of Quantum Chemistry		QCEIMS, CFM-EI, comparison between first principle simulation and expert system
FragHub: A Mass Spectral Library Data Integration Workflow	Workflow	Analytical Chemistry		FragHub, mass spectral lib integration
Evaluation of the performance of a tandem mass spectral library with mass spectral data extracted from literature	Database	Drug Testing and Analysis		MSforID, mass spectrum analysis
Comparative Evaluation of Electron Ionization Mass Spectral Prediction Methods	Quantum Chemistry, Machine Learning, Algorithm			QCEIMS, CFM-EI, NEIMS comparison
Computational mass spectrometry for small-molecule fragmentation	General models of fragmentation, Simulation Software, Fragmentation Prediction Software, Machine Learning, Heuristic Method, Combination of MetFrag and spectral library search, Classifier, Kernel-based method, Combinatorial Optimization Model	TrAC Trends in Analytical Chemistry		DENDRAL, Mass Frontier 4, Mass Frontier 6, ACD Fragmenter, ISIS, MetFrag, MetFusion, Varmuza feature-based classification approach, Heinonen et al. kernel-based approach, Fragmentation Trees, mass spectrometry fragmentation, review paper
Searching molecular structure databases using tandem MS data: are we there yet?	Automated Method, Machine Learning	Current Opinion in Chemical Biology		CFM-ID, MetFrag, MAGMa, FingerID, CSI:FingerID, MetFrag2.2, MAGMa+, MSFINDER, MIDAS, IOKR version of CSI:FingerID, metabolite identification, review paper
Unsupervised machine learning for exploratory data analysis in imaging mass spectrometry	Unsupervised machine learning	Mass Spectrometry Reviews		PCA, IMS data analysis, Project Link: BioMap, review paper
Machine Learning in Small-Molecule Mass Spectrometry	GNN, Transformer, Machine learning	Annual Review of Analytical Chemistry
NIST20	data library			project website, purchase required
NPLIB1	dataset,benchmark			data source, deriving from CANOPUS by filtering to [M+H] spectra
GeMS	dataset			dataset collected by Dreams, LC-MS, 714 million MS/MS spectra
MSnLib: efficient generation of open multi-stage fragmentation mass spectral libraries	Spectral Library	Nature Methods		MSnLib

1.5 Spectrum Databases (API Usage)

Name & Link	Request Type	Description	Return information	API_link
GNPS	html	This gets you all spectra but without peaks.	Spectrum Metadata	json
MoNA	curl	Search for Mass Spectrum

2. Mass Spectrometry (Peptides)

2.1 Forward Task (Peptides → Spectrum)

Computational methods for predicting peptides mass spectra

Paper Title & Link	Method Type	Venue	Code	Notes
Full-Spectrum Prediction of Peptides Tandem Mass Spectra using Deep Neural Network	CNN	Analytical Chemistry 2020		PredFull
Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning	Transformer	Nature Methods 2019		Prosit
High-quality MS/MS spectrum prediction for data-dependent and data-independent acquisition data analysis	LSTM	Nature Methods 2019		DeepMass:Prism

2.2 Inverse Task (Spectrum → Peptides)

AI approaches for peptides identification and quantification

Paper Title & Link	Method Type	Venue	Code	Notes
InstaNovo enables diffusion-powered de novo peptide sequencing in large-scale proteomics experiments	Transformer, diffusion model	nature machine intelligence		InstaNovo
De Novo Mass Spectrometry Peptide Sequencing with a Transformer Model	Transformer	ICML 2022		CasaNovo
Bridging the Gap between Database Search and De Novo Peptide Sequencing with SearchNovo	Transformer, Database	ICLR 2025	—	SearchNovo
De novo peptide sequencing by deep learning	CNN, LSTM	PNAs 2017		deepNovo
Computationally Instrument-Resolution-independent De Novo Peptide Sequencing for High-Resolution Devices	CNN, LSTM	Nature Machine Intelligence 2021	—	PointNovo
AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information	Transformer	NeurIPS 2024		AdaNovo
Accurate de novo peptide sequencing using fully convolutional neural networks	CNN	Nature Communications 2023		PepNet
A transformer model for de novo sequencing of data-independent acquisition mass spectrometry data	Transformer	Nature Method 2025		cascadia
Towards highly sensitive deep learning-based end-to-end database search for tandem mass spectrometry	Transformer	Nature machine intelligence 2025		Deepsearch, database search
π-PrimeNovo: an accurate and efficient nonautoregressive deep learning model for de novo peptide sequencing	Transformer	Nature Communications 2025		π-PrimeNovo

2.3 Datasets, Benchmark and Review

Paper Title & Link	Method Type	Venue	Code	Notes
AlphaPeptDeep: a modular deep learning framework to predict peptide properties for proteomics	Benchmark	Nature communications		AlphaPeptDeep
NovoBench: Benchmarking Deep Learning-based De Novo Peptide Sequencing Methods in Proteomics	Benchmark	NeurIPS 2024		NovoBench

2.4 General Tools

Paper Title & Link	Feasible scene	Venue	Code	Notes
MSBooster: improving peptide identification rates using deep learning-based features	Rating between LC-MS/MS spectra and Piptide	Nature communications		python package

---

3. NMR Spectroscopy (Small Molecules)

3.1 Forward Task (Molecule → Spectrum)

Paper Title & Link	Method Type	Venue/Year	Data Source	Metric	Code / Data	Notes
Enhancing NMR Shielding Predictions of Atoms-in-Molecules Machine Learning Models with Neighborhood-Informed Representations	-	preprint	^1H/^13C NMR spectra	-	—
Toward a unified benchmark and framework for deep learning-based prediction of nuclear magnetic resonance chemical shifts	Transformer	Nature Computional Science 2025	3Dstructure to NMR		Zenodo
Deep Learning Network for NMR Spectra Reconstruction in Time-Frequency Domain and Quality Assessment	Transformer + CNN hybrid	Nature Communications 2025	1D/2D NMR time-series	RMSE (dB)	SpectraNet · Zenodo	Time-frequency domain reconstruction; spectral quality scoring
GT-NMR: a novel graph transformer-based approach for accurate prediction of NMR chemical shifts	Graph + Transformer				GT-NMR ·
PROSPRE: Solvent-aware ^1H NMR chemical shift prediction using deep learning	Deep Learning + Solvent-aware	Metabolites 2024	Experimental NMR data	MAE < 0.10 ppm	—	Trained on large-scale solvent-specific datasets
Prediction of chemical shift in NMR: A review	Review	—	—	—	—	Survey of methods & datasets
iShiftML: Highly Accurate Prediction of NMR Chemical Shifts	Hybrid ML + QM	—	QM descriptors	< 0.2 ppm	—	Fast inference; QM features required
TransPeakNet: Solvent-Aware 2D NMR Prediction via Multi-Task Pre-Training and Unsupervised Learning	GNN + Multitask	Chem paper 2023	Graph-based input	2D spectra	Code · Data	1D→2D solvent-aware prediction
NMR shift prediction from small data quantities	ML	—	NMRShiftDB2	MAE (ppm)	—	Small-data learning
NMR-spectrum prediction for dynamic molecules	ML-Dynamics	—	Simulated ensembles	Time-avg ppm	—	Conformational averaging
Machine learning in NMR spectroscopy	Review (DL)	—	NMRShiftDB2	—	—	Multitask trends & outlook
A framework for automated structure elucidation from routine NMR spectra	ML-based structure elucidation	Chem. Sci. 2021	1D ^1H/^13C NMR spectra	Top-10 accuracy	—	ML framework for automated structure elucidation
Deep learning enabled ultra-high quality NMR chemical shift prediction from spin echo spectra	Deep Learning + Signal Processing	Science 2024	Spin echo NMR spectra	MAE (ppm)	—	High-resolution chemical shift prediction

---

3.2 Inverse Task (Spectrum → Molecule)

Paper Title & Link	Method Type	Venue/Year	Input	Metric	Code / Data	Notes
Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra	-	openreview	^1H/^13C NMR spectra	-	—	DiT Based 3D molecular generation
NMRMind: A Transformer-Based Model Enabling the Elucidation from Multidimensional NMR to Structures	Transformer	Analytical Chemistry 2025	^1H/^13C NMR spectra	Top-k accuracy	—
DIFFNMR: Diffusion Models for Nuclear Magnetic Resonance Spectra Elucidation		Arxiv 2025	^1H/^13C NMR spectra	Top-k accuracy	—	Diffusion Based Generation
A Transformer Based Generative Chemical Language AI Model for Structural Elucidation of Organic Compounds	Transformer-based Generative AI	Arxiv 2024	IR, UV, ^1H NMR spectra	Top-15 accuracy	—	End-to-end structure elucidation via chemical language modeling
NMR-Solver: Automated Structure Elucidation via Large-Scale Spectral Matching and Physics-Guided Fragment Optimization	Hybrid Spectral Matching + Physics-guided Optimization	Arxiv 2025	^1H/^13C NMR spectra	Top-1 accuracy	—	Combines spectral matching with physics-guided optimization
Accurate and Efficient Structure Elucidation from Routine One-Dimensional NMR Spectra Using Multitask Machine Learning	CNN + Transformer (multitask)	Arxiv 2024	1D spectra	Top-1 / Top-k	—	High accuracy with 1D only
Learning the Language of NMR Structure Elucidation from NMR Spectra Using Transformer Models	Transformer (sequence modeling)	ChemRxiv 2023	1D spectra	—	—	Treats NMR as a language; structure reasoning
DeepSAT: Learning Molecular Structures from NMR Data	Multimodal DL	—	NMR spectra	Structure accuracy	—	Uses NPAtlas, NPASS, GNPS etc.; multimodal molecular structure learning
Bayesian approach to structural elucidation with crystalline-state SSNMR	Bayesian / probabilistic	arXiv 2019	Solid-state NMR	Top-5	—	Requires crystal info
Deep RL + GCN for NMR inverse problem	RL (MCTS) + GCN	J. Phys. Chem. Lett. 2022	Shift table	Top-3	—	Effective for small molecules

3.3 Dataset

Name & Link	Type	Venue/Year	Size / Modality	Real / Sim	Download / Access
Multimodal Spectroscopic Dataset (includes NMR)	Benchmark (multi-spectra)	NeurIPS 2024	7.9e5+ synthetic spectra	Sim	Zenodo (via paper)
IR–NMR Multimodal Computational Spectra Dataset	Dataset (IR + NMR)	Nature Sci. Data 2025	177,461 spectra	Sim (MD + DFT + ML)	Zenodo
NMRShiftDB2	Database (small molecules)	—	~50k; ¹H/¹³C	Real	Open
BMRB	Database (bio-molecules)	—	>13k biomolecules; ¹H/¹³C/¹⁵N…	Real	FTP/STAR
SDBS	Database (multi-modal)	—	~14k; ¹H/¹³C/IR/MS/UV	Real	Crawl/script
2DNMRGym (HSQC)	Simulated 2D dataset	2024	22k+ HSQC	Sim	HuggingFace/Zenodo
NMRMixDB	Mixtures	—	~3k; ¹H	Real	Open
HMDB 5.0	Database (metabolites)	—	4,149; ¹H/¹³C	Real	Open-source
BMRB20 (metabolites subset)	Database (metabolites)	—	1,200+; ¹H/¹³C	Real	Open-source
NMRShiftDB2 (updated)	Database (small molecules)	—	53,954; ¹H/¹³C	Real	Open-source
SDBS23	Database (natural products)	—	15,218 (¹H) / 13,457 (¹³C); 900+ NPs	Real	Open-source
NP-MRD	Database (natural products)	—	1,290; ¹H/¹³C	Real	Open-source
NAPROC-13	Database (natural products)	—	6,000+; ¹³C	Real	Open-source
CH-NMR-NP	Database (natural products)	—	30,500; 926 compounds; ¹H/¹³C	Real	Open-source
Spektraris-NMR	Specialized subset	—	466 samples; 250 taxanes; ¹H/¹³C	Real	Open-source
C6H6	Database	—	506 entries; ¹H/¹³C	Real	Open-source
Ilm-NMR-P31	Element-specific database	—	14,250 (31P spectra) / 13,730 entries	Real	Open-source
KnowItAll NMR	Commercial database	—	1.28 M+ spectra; ¹H/¹³C	Real (Commercial)	Commercial (Wiley)
Micronmr	Commercial database	—	1.0 M+ spectra; ¹³C	Real (Commercial)	Commercial
NMRBank	Open repository	—	225,809 (¹H) / 149,135 (¹³C)	Real	Open-source

                                 |

---

3.4 Clustering & Representation Learning for NMR

Paper Title & Link	Method	Venue/Year	Task	Notes
Statistical HOmogeneous Cluster SpectroscopY (SHOCSY) (Anal. Chem. 2014)	Statistical clustering	2014	¹H metabolomics clustering	Classic baseline
Sparse Convex Wavelet Clustering (includes NMR signals)	Wavelet + convex clustering	arXiv	Signal clustering	Joint denoising + clustering
Deep representation learning for NMR spectral clustering	Autoencoder / DL	—	Spectral clustering	DL embedding → K-means/HC

---

3.5 Foundation Models & Chemistry LMs related to NMR

Model / Paper	Family	Venue/Year	Link	Notes
ChemGPT	Large chemistry LM	—	—	Molecular generative & reasoning; spectra-to-structure transfer
DETANet	DL architecture	—	—	Chemistry perception; potential for spectra-conditioned tasks
DreaMS (MS foundation model)	Transformer (spectra FM)	Nat. Biotech. 2025	GitHub	Cross-modal FM idea transferrable to NMR

📚 Synthetic NMR Papers & Datasets

Category	Name	What it offers	Typical Use	Link
Dataset & Pipeline	ARTINA – 100‑Protein NMR Dataset	1329 2D–4D spectra + chemical shift assignments + protein structures	Train/test protein NMR auto‑assignment & structure inference	https://www.nature.com/articles/s41597-023-02879-5
Database (LLM‑extracted)	NMRBank	~225k small‑molecule records (SMILES, ¹H/¹³C shifts)	Build ML models for shift/spectrum prediction	https://pmc.ncbi.nlm.nih.gov/articles/PMC12118362/
Synthetic multimodal	IR‑NMR Dataset	177k IR spectra + 1.2k NMR shifts (DFT+ML)	Cross‑modal learning / pretraining	https://www.nature.com/articles/s41597-025-05729-8
Carbohydrates	GlycoNMR	2,609 glycans with 211k NMR shifts	Domain‑specific ML (carbohydrates)	https://arxiv.org/abs/2311.17134
2D spectra	2DNMRGym	22k+ HSQC spectra + SMILES (partly human‑annotated)	Train/benchmark 2D NMR predictors	https://arxiv.org/abs/2505.18181

---

🛠️ Tools for Synthetic NMR Generation

Tool	What it does	How to use	Required Inputs	Output	Link
MestReNova	Comprehensive NMR data analysis software for processing 1D/2D spectra, including peak picking, integration, and chemical shift assignment	Process NMR spectra, analyze data, and assign chemical shifts using MestReNova interface	NMR spectra: 1D, 2D, 3D; chemical shift (ppm), coupling constants (Hz), integration, peak assignment	Processed spectra, chemical shift assignments, peak integration	MestReNova Official Website
Spinach (Matlab)	Physics‑based simulation of 1D/2D spectra (COSY, HSQC, NOESY), relaxation, MAS NMR	Define spin system + select pulse sequence + run simulation in Matlab	Spin system: isotopes, chemical shifts (ppm), J couplings (Hz), CSA/dipolar terms; external field (B₀); pulse sequence	Time‑domain FID & frequency‑domain spectra (1D/2D)	https://en.wikipedia.org/wiki/Spinach_(software)
ORCA / NWChem	Quantum chemistry calculation of shielding tensors and J couplings	Run DFT/ab initio with NMR keyword	3D geometry (XYZ/MOL/PDB), basis set, method, optional solvent model	Isotropic shieldings (σ), J‑couplings, CSA tensors	https://github.com/nwchemgit/nwchem
NMRDB	Online prediction of ¹H/¹³C (1D and 2D COSY/HSQC/HMBC)	Draw molecule or paste SMILES on web UI	Molecular structure (drawn or SMILES), optional solvent/field strength	Simulated 1D/2D spectra, JCAMP export	https://www.nmrdb.org/
ChemAxon NMR Predictor	Predicts ¹H/¹³C chemical shifts and spectra (GUI + CLI)	Use MarvinSketch or cxcalc nmr CLI	Structure input (SMILES, MOL, SDF), optional solvent/field	Chemical shifts, spectra, JCAMP files	https://docs.chemaxon.com/display/docs/NMR%2BPredictor
NMRbox	VM platform bundling many NMR tools (TopSpin, Sparky, CCPN, etc.)	Launch VM, import spectra, run pipelines	Experimental or synthetic spectra, peak lists	Processed spectra, assignments, structural analysis	https://nmrbox.nmrhub.org/software

---

4. IR Spectroscopy (Small Molecules)

4.1 Forward Task (Molecule → Spectrum)

Infrared spectrum prediction from molecular structures

Paper Title & Link	Venue	Method Type	Input	Data Source	Code	CKPT
Machine Learning Molecular Dynamics for the Simulation of Infrared Spectra (2017)	Chem. Sci.	ML (HDNNP + NN Dipole)	3D coordinates	—	—	—
A Machine Learning Protocol for Predicting Protein Infrared Spectra (2020)	J. Am. Chem. Soc.	ML (MLP)	3D coordinates	—	—	—
Predicting Infrared Spectra with Message Passing Neural Networks (2021)	J. Chem. Inf. Model.	DL (MPNN + FFNN)	Simulated: SMILES Experimental: SMILES+Phase	Simulated: PubChem Experimental: NIST, PNNL, AIST, Coblentz
Graphormer-IR: Graph Transformers Predict Experimental IR Spectra (2024)	J. Chem. Inf. Model.	DL (Graph Transformer + MLP, 1D-CNN)	SMILES+Phase	NIST, AIST, Coblentz
Neural Network Approach for Predicting Infrared Spectra from 3D Molecular Structure (2024)	Chem. Phys. Lett.	DL (MPNN)	3D Molecular Structure	NIST
Infrared Spectra Prediction for Functional Group Region Utilizing a Machine Learning Approach with Structural Neighboring Mechanism (2024)	Anal. Chem.	ML	SMILES	CIAD		--
Infrared Spectra Prediction Using Attention-Based Graph Neural Networks (2024)	Digital Discovery	DL (GNN)	SMILES	NIST		—
Prediction of the Infrared Absorbance Intensities and Frequencies of Hydrocarbons: A Message Passing Neural Network Approach (2024)	J. Phys. Chem. A	DL (MPNN + FFNN)	SMILES	GDB
Unlocking the Potential of Machine Learning in Enhancing Quantum Chemical Calculations for Infrared Spectral Prediction (2025)	ACS Omega	ML (Multioutput Regressor + Random Forest Regressor)	3D Molecular Structure	Gaussian 16	Supporting Information	—

4.2 Inverse Task (Spectrum → Molecule)

Molecular characterization from infrared spectra

Paper Title & Link	Venue	Method Type	Output	Data Source	Code	CKPT
Can LLMs Solve Molecule Puzzles? A Multimodal Benchmark for Molecular Structure Elucidation (2024)	NeurIPS2024	LLM (GPT‑4o, Claude‑3, Gemini, etc.)	SMILES	—		—
Leveraging Infrared Spectroscopy for Automated Structure Elucidation (2024)	Commun. Chem.	DL (Transformer)	SMILES	Simulated: PubChem Experimental: NIST
Transformer-Based Models for Predicting Molecular Structures from Infrared Spectra Using Patch-Based Self-Attention (2025)	J. Phys. Chem. A	DL (Transformer + Patch-based Self-Attention)	SMILES	Simulated: PubChem, QM9S Experimental: NIST
Revolutionizing Spectroscopic Analysis Using Sequence-to-Sequence Models I: From Infrared Spectra to Molecular Structures (2025)	ChemRxiv	DL (GRU, LSTM, GPT, Transformer)	SELFIES	QM9, PC9	—	—
Setting new benchmarks in AI‑driven infrared structure elucidation (2025)	Digit. Discov.	DL (Transformer + Patch‑based Self‑Attention)	SMILES	Simulated: IRtoMol, Multimodal Spectroscopic Dataset Experimental: NIST		GitHub

4.3 IR Datasets

Dataset / Method Name & Link	Size	Data Source	Real / Simulated	Element Coverage
Chemprop-IR	85,232	PubChem (SMILES for molecular structures)	Simulated (GFN2-xTB)	C, H, O, N, S, P, Si, F, Cl, Br, I
CMPNN	31,570	GDB (SMILES for molecular structures)	Simulated (DFT)	C, H
Multimodal Spectroscopic Dataset	794,403	USPTO reaction dataset (SMILES for molecular structures)	Simulated (MD)	C, H, O, N, S, P, Si, B, F, Cl, Br, I
IRtoMol	634,585	PubChem (SMILES for molecular structures)	Simulated (MD + PCFF forcefield)	C, H, O, N, S, P, F, Cl, Br, I
MolPuzzle	216 (Picture format)	—	Mixed	—
IR–NMR Multimodal Computational Spectra Dataset	177,461	USPTO (SMILES for molecular structures)	Simulated (MD + DFT + ML)	C, H, O, N, S, P, Si, B, F, Cl, Br, I
QM9S	133,885	QM9 (re-optimized geometries)	Simulated (DFT)	C, H, O, N, F
SRD 35	5,228 (gas-phase)	—	Real	—
NIST Chemistry WebBook	>16,000	—	Real	—
Coblentz Society Spectral Database	>9,500	—	Real	—
NWIR	~1,000–1,500 (gas-phase)	—	Real	—
AIST SDBS	~54,100	—	Real	—

4.4 General Tools

Tool & Link	Type	Main Functions / Uses
RDKit	Open-source library (Python/C++)	Molecular descriptors for IR–structure fusion, Morgan/ECFP fingerprints for IR-guided search, SMARTS substructure from IR functional-group hints, Conformers + MMFF/UFF for vibrational context, InChI/SMILES I/O to link spectra and structures
Psi4	Open-source library (Python/C++)	High‑accuracy, open‑source ab initio quantum chemistry suite for molecular properties
GFN2-xTB	Open-source semiempirical QM (Fortran/Python)	Fast IR frequency estimation, IR peak-to-structure attribution, Conformer effects on IR bands, Batch geometry preparation, Solvent-adjusted IR shifts
cclib	Open-source library (Python)	Provides parsers for output files of computational chemistry packages
SELFIES	Open-source library (Python)	SMILES↔SELFIES conversion
Open Babel	Open-source chemistry toolbox (C++/Python)	Format conversion, 3D generation and conformer search (MMFF94/UFF), SMARTS substructure and fingerprints (FP2/FP3/MACCS)
Gaussian 16	Commercial quantum chemistry software	DFT IR frequencies/intensities, Harmonic and anharmonic IR (VPT2), Normal‑mode peak assignment, Solvent models (PCM/SMD) for IR
ChemCraft	Commercial visualization software	Visualize vibrational spectra with Lorentzian/Gaussian broadening, Compare calculated vs. experimental spectra, View MOs/SCF graphs, Parse outputs from Gaussian/ORCA/GAMESS/NWChem etc.
Q‑Chem	Commercial quantum chemistry software	DFT IR frequencies/intensities, Harmonic and anharmonic IR (VPT2/VCI/TOSH), Normal‑mode and Raman analysis

---

5. Multimodal Spectroscopy (Small Molecules)

5.1 Forward Task (Molecule → Multiple Spectra)

Joint prediction of multiple spectral modalities from molecular structures

Paper Title & Link	Modalities	Method Type	Venue/Year	Code	Notes
Unraveling Molecular Structure: A Multimodal Spectroscopic Dataset for Chemistry	IR + MS + ^1H/^13C/HSQC-NMR (simulated)	Dataset + baselines (CNN/XGBoost)	NeurIPS D&B 2024		Provides ~790k multimodal spectra and forward-prediction baselines to pretrain joint predictors.
IR–NMR Multimodal Computational Spectra Dataset	IR + NMR (simulated)	Dataset (MD/DFT/ML)	Sci. Data 2025	—	Curated paired IR–NMR set enabling multi-task forward models and cross-modal transfer.

5.2 Inverse Task (Multiple Spectra → Molecule)

Multimodal integration for enhanced molecular identification

Paper Title & Link	Modalities	Method Type	Venue/Year	Code	Notes
A Transformer-Based Generative Chemical Language AI Model for Structural Elucidation of Organic Compounds	IR + UV/Vis + ^1H NMR	Encoder–decoder Transformer (generative CASE)	J. Cheminformatics 2025	—	End-to-end spectra→structure generation; strong single-paper reference for joint use of IR/UV/NMR.
Leveraging Infrared Spectroscopy for Automated Structure Elucidation	IR	Transformer	Commun. Chem. 2024		Single-modality IR baseline frequently paired with NMR/MS to form multimodal pipelines.

5.3 Representation Learning & Fusion (Model-Agnostic Methods used for Spectra)

Paper Title & Link	What it Brings	Method Type	Venue/Year	Code	Notes
Ming-Omni: A Unified Multimodal Model for Perception and Generation	Unified encoder/decoders; modality-specific MoE routers	MoE-UMM (image/text/audio/video)	arXiv 2025		Useful design for preventing cross-modal conflicts when adding spectral modalities as new “experts”.
Boosting Multimodal Learning via Disentangled Gradient Learning (DGL)	Decouples encoder vs. fusion optimization to avoid gradient interference	Training framework	arXiv 2025		Practical recipe when IR/NMR/MS encoders underperform once fused.
I2MoE: Interpretable Multimodal Interaction-aware Mixture-of-Experts	Interaction-specialized experts + sample-level explanations	MoE fusion	arXiv/ICML 2025		Lets you see which modality interactions (e.g., IR×NMR) drive a prediction.
MM-Embed: Universal Multimodal Retrieval with MLLMs	Bi-encoder + MLLM re-ranker for flexible text↔image retrieval	Multimodal embeddings	ICLR 2025	HF Card	Template for spectra/text retrieval (e.g., query by IR peaks + structural hints).
SpecEmbedding: Supervised Contrastive Learning Leads to More Reasonable Spectral Embeddings	MS spectral encoder with supervised contrastive learning	Transformer encoder + SupCon	Analytical Chemistry ASAP 2025	· Demo	High-quality MS embeddings usable as one branch in IR+NMR+MS fusion.
UAE: Can Understanding and Generation Truly Benefit Together — or Just Coexist?	Unified auto-encoder view (I2T ↔ T2I) with RL to align understanding & generation	Unified multimodal learning (UMM)	arXiv 2025		Useful recipe for tying inverse (spectra→structure) with forward (structure→spectra) via reconstruction.

5.4 Datasets & Benchmarks (Multimodal, Small Molecules)

Name & Link	Modalities	Type	Size	Real / Simulated	Website / Code	Notes
Multimodal Spectroscopic Dataset (MSD)	IR + MS + ^1H/^13C/HSQC-NMR	Benchmark + Dataset	~790k molecules	Simulated	Site · Zenodo · GitHub	Common pretraining/eval source for multimodal spectral models.
IR–NMR Multimodal Computational Spectra Dataset	IR + NMR	Dataset	177,461 spectra	Simulated	Article · Zenodo	Paired IR–NMR for cross-modal learning and evaluation.
SDBS (AIST)	IR + MS + ^1H/^13C-NMR + Raman + ESR	Database	~34k molecules	Real	SDBS	Broad real-world spectra across modalities for validation.
NIST Chemistry WebBook / SRD-35	IR (gas-phase) + MS + UV/Vis	Database	5,228 IR (SRD-35); 16k+ IR & 33k+ MS overall	Real	WebBook · SRD-35	Standard references and downloadable IR libraries.
SimXRD-4M (related)	Powder XRD	Benchmark + Dataset	4.06M patterns	Simulated	OpenReview	Not small-molecule spectra, but relevant template for large-scale physics-faithful simulation & benchmarking.

---

6. X-ray Diffraction (XRD) (Crystals)

6.1 Forward Task (Crystal → Pattern)

Prediction of XRD patterns from crystal structures

6.2 Inverse Task (Pattern → Crystal)

Crystal structure determination from XRD patterns

---

License

📄 This project is licensed under the MIT License — see the LICENSE file for details.

---