Multimodal deep learning from spectra for small-molecule structure identification: enhancing robustness with mixed-condition training

• URL: http://arxiv.org/abs/2609.14360v1
• Date: Sun, 13 Sep 2026 07:39:56 GMT
• Title: Multimodal deep learning from spectra for small-molecule structure identification: enhancing robustness with mixed-condition training
• Abstract summary: We incorporate domain knowledge from spectroscopy and chemistry into mixed-condition training for candidate structure reranking.<n>The protocol incorporates perturbations tailored to each spectroscopic modality and chemically informed spectrum replacements to cover variations in spectral availability, quality, and consistency.
• Score: 11.492759216075974
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In practical molecular characterization, small-molecule structure identification benefits from complementary spectroscopic evidence, but missing, degraded, or mismatched spectra challenge multimodal models. Herein, we incorporate domain knowledge from spectroscopy and chemistry into mixed-condition training for candidate structure reranking, using a reproducible evaluation protocol and mixture-of-experts (MoE) fusion. The protocol incorporates perturbations tailored to each spectroscopic modality and chemically informed spectrum replacements to cover variations in spectral availability, quality, and consistency. A total of 79,462 test samples were evaluated across 30 predefined conditions using simulated spectra from the Multimodal Spectroscopic Dataset (MSSD) for mass spectrometry (MS), infrared (IR) spectroscopy, and 1H and 13C nuclear magnetic resonance (NMR), with up to 128 hard candidate structures per sample. A controlled two-by-two factorial comparison of complete-input training versus mixed-condition training and vanilla concatenation versus MoE fusion, with matched evaluation conditions, showed that mixed-condition training provided the main gains in both architectures. For MoE, mean reciprocal rank (MRR), averaged equally across conditions, increased from 0.9203 to 0.9763, a relative increase of 6.08%. Recall at rank 1 (R@1), averaged over the same conditions, increased from 89.50% to 96.36%, an increase of 6.86 percentage points and a relative increase of 7.67%. IR-only and MS/MS-only MRR increased from 0.4337 to 0.9307 and from 0.3711 to 0.8575, reaching 2.15 and 2.31 times their respective baseline values, while complete-input performance remained high. These results support integrating domain knowledge into training-condition design to improve robustness, with further gains from MoE under mixed-condition training.

Related papers

• CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation [49.0564416580726]
  We distill the compositional judgments of a cross-attentive reranker into an embedding model.<n>We show that CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points.<n>The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.
  arXiv  Detail & Related papers  (2026-09-03T16:50:29Z)
• Hypothesis-and-Refinement Learning of Organic Structures from Multimodal Spectroscopic Data [41.531262858699925]
  We develop a scalable hypothesis-refinement paradigm that tightly integrates spectral evidence with large-scale molecular priors.<n>We introduce textbfSpectroMol, a spectrum-to-structure model that proposes chemically valid molecular hypotheses conditioned on multimodal spectral inputs.<n>We also develop textbfMS-Mol2Mol, a high-resolution mass-constrained molecular generator that integrates molecular formula, exact mass, and degree of unsaturation.
  arXiv  Detail & Related papers  (2026-07-22T06:50:53Z)
• SCENT: Aligning Mass Spectra with Molecular Structure for Olfactory Perception [48.96518000981492]
  We explore direct electron ionization mass spectrometry (EI-MS) as an alternative input modality for olfactory prediction.<n>We contribute Spectrum-to-Chemical Embedding alignmeNT (SCENT), a multi-modal contrastive learning framework that aligns EI-MS representations with pretrained chemical structure embeddings.<n>On the multi-label odor descriptor prediction task, SCENT significantly outperforms MS-only baselines and achieves olfactory performance comparable to structure-based models.
  arXiv  Detail & Related papers  (2026-05-26T13:28:57Z)
• FRIGID: Scaling Diffusion-Based Molecular Generation from Mass Spectra at Training and Inference Time [52.735012862324766]
  We present FRIGID, a framework with a novel diffusion language model that generates structures conditioned on mass spectra.<n>We show how forward fragmentation models enable inference-time scaling by identifying spectrum-inconsistent fragments.<n>Further empirical analyses show that FRIGID exhibits log-linear performance scaling with increasing inference-time compute.
  arXiv  Detail & Related papers  (2026-04-17T19:11:18Z)
• Spectral Coherence Index: A Model-Free Metric for Protein Structural Ensemble Quality Assessment [27.622823036293525]
  We evaluate the Spectral Coherence Index (SCI), a model-free, rotation-invariant summary.<n>SCI is most useful when embedded in a multimetric QC workflow for heterogeneous protein ensembles.
  arXiv  Detail & Related papers  (2026-03-26T20:09:41Z)
• From Static Spectra to Operando Infrared Dynamics: Physics Informed Flow Modeling and a Benchmark [67.29937933325849]
  Operando IR Prediction aims to forecast the time-resolved evolution of spectral fingerprints'' from a single static spectrum.<n>OpIRSpec-7K comprises 7,118 high-quality samples across 10 distinct battery systems.<n>ABCC significantly outperforms state-of-the-art static, sequential, and generative baselines.
  arXiv  Detail & Related papers  (2026-02-20T18:58:43Z)
• NMRTrans: Structure Elucidation from Experimental NMR Spectra via Set Transformers [41.6373573055135]
  We build NMRSpec, a large-scale corpus of experimental $1$H and $13$C spectra mined from chemical literature.<n>We propose NMRTrans, which models spectra as unordered peak sets and aligns the model's inductive bias with the physical nature of NMR.
  arXiv  Detail & Related papers  (2026-02-10T03:37:41Z)
• How well can off-the-shelf LLMs elucidate molecular structures from mass spectra using chain-of-thought reasoning? [51.286853421822705]
  Large language models (LLMs) have shown promise for reasoning-intensive scientific tasks, but their capability for chemical interpretation is still unclear.<n>We introduce a Chain-of-Thought (CoT) prompting framework and benchmark that evaluate how LLMs reason about mass spectral data to predict molecular structures.<n>Our evaluation across metrics of SMILES validity, formula consistency, and structural similarity reveals that while LLMs can produce syntactically valid and partially plausible structures, they fail to achieve chemical accuracy or link reasoning to correct molecular predictions.
  arXiv  Detail & Related papers  (2026-01-09T20:08:42Z)
• Breaking the Modality Barrier: Generative Modeling for Accurate Molecule Retrieval from Mass Spectra [60.08608779794957]
  We propose GLMR, a Generative Language Model-based Retrieval framework.<n>In the pre-retrieval stage, a contrastive learning-based model identifies top candidate molecules as contextual priors for the input mass spectrum.<n>In the generative retrieval stage, these candidate molecules are integrated with the input mass spectrum to guide a generative model in producing refined molecular structures.
  arXiv  Detail & Related papers  (2025-11-09T07:25:53Z)
• DiffSpectra: Molecular Structure Elucidation from Spectra using Diffusion Models [68.19129717255053]
  We present DiffSpectra, a generative framework that formulates molecular structure elucidation as a conditional generation process.<n>Our experiments demonstrate that DiffSpectra accurately elucidates molecular structures, achieving 40.76% top-1 and 99.49% top-10 accuracy.
  arXiv  Detail & Related papers  (2025-07-09T13:57:20Z)
• TransPeakNet: Solvent-Aware 2D NMR Prediction via Multi-Task Pre-Training and Unsupervised Learning [5.7279868722119325]
  We introduce an unsupervised training framework for predicting cross-peaks in 2D NMR.<n>Our approach pretrains an ML model on an annotated 1D dataset of 1H and 13C shifts, then finetunes it in an unsupervised manner.<n> Evaluation on 479 expert-annotated HSQC spectra demonstrates our model's superiority over traditional methods.
  arXiv  Detail & Related papers  (2024-03-17T21:52:51Z)
• AdaNovo: Adaptive \emph{De Novo} Peptide Sequencing with Conditional Mutual Information [46.23980841020632]
  We propose AdaNovo, a novel framework that calculates conditional mutual information (CMI) between the spectrum and each amino acid/peptide. AdaNovo excels in identifying amino acids with post-translational modifications (PTMs) and exhibits robustness against data noise.
  arXiv  Detail & Related papers  (2024-03-09T11:54:58Z)
• A Multisensor Hyperspectral Benchmark Dataset For Unmixing of Intimate Mixtures [3.0314808581426846]
  We generate a comprehensive ground truth dataset of intimately mixed mineral powders. Thirteen different hyperspectral sensors have been used to acquire the reflectance spectra of these mixtures. We believe these data will be beneficial for validating advanced methods for nonlinear unmixing and material composition estimation.
  arXiv  Detail & Related papers  (2023-08-30T11:48:36Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.