Nearest-neighbour baselines for fingerprint prediction from MS/MS spectra under different assumptions

Ling Min Serena Khoo

cs.LGSep 30, 2026

Nearest-neighbour baselines for fingerprint prediction from MS/MS spectra under different assumptions

Authors: Ling Min Serena Khoo

Organizations: MIT

Abstract

It has recently been shown that nearest-neighbour retrieval provides a strong baseline for molecular fingerprint prediction from MS/MS spectra, with several variants matching or outperforming current deep learning models (Khoo and Barzilay, 2026; Liu et al., 2026; Gupta et al., 2026). Importantly, "nearest neighbour" encompasses a family of retrieval methods that differ in the information assumed to be available at inference. In this report, we systematically compare several nearest-neighbour variants and show how these differing assumptions affect performance. Our goal is to establish stricter baselines that enable more rigorous benchmarking and better measure progress in this area.

Figures & tables

		NPLIB1		MassSpecGym
Assumption	Similarity	Scaffold	Random	Scaffold	Random
(1) No chemical formula	Cosine	0.208	0.519	0.235	0.740
	Modified cosine	0.236	0.517	0.246	0.700
	Neutral-loss cosine	0.187	0.533	0.220	0.749
	Cosine Similarity (DreaMS)	0.263	0.691	0.299	0.895
	Binned cosine	0.178	0.592	0.231	0.850

Table 1: Binary Jaccard index (Tanimoto similarity) between the predicted and the true fingerprint (Morgan, radius 2, 4096 bits). Predictions are binarised at 0.5. Mean over the full test split. + sub-formula : same-formula retrieval with the given similarity, sub-formula rule for test spectra without a same-formula training spectrum. Best value per column in bold.

		NPLIB1		MassSpecGym
Assumption	Similarity	Scaffold	Random	Scaffold	Random
(1) No chemical formula	Cosine	0.313	0.593	0.331	0.777
	Modified cosine	0.346	0.598	0.347	0.745
	Neutral-loss cosine	0.290	0.602	0.311	0.781
	Cosine Similarity (DreaMS)	0.380	0.754	0.400	0.913
	Binned cosine	0.276	0.650	0.315	0.869

Table 2: Cosine similarity between the raw prediction (the retrieved fingerprint, the sub-formula consensus before thresholding) and the true fingerprint. Mean over the full test split. + sub-formula : same-formula retrieval with the given similarity, sub-formula rule for test spectra without a same-formula training spectrum. Best value per column in bold.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Split	Train	Val	Test	Coverage	Leakage	FP oracle
NPLIB1	Scaffold	12,909	2,688	2,689	38.2%	1.6%	0.523
NPLIB1	Random	12,800	2,742	2,744	80.7%	62.9%	0.870
MassSpecGym	Scaffold	76,243	16,041	16,042	41.0%	14.3%	0.562
MassSpecGym	Random	75,828	16,248	16,250	97.1%	95.3%	0.977

Appendix

Table 3: Datasets and splits. Coverage : test spectra whose neutral formula occurs in the training split (the spectra that same-formula retrieval can serve). Leakage : test spectra whose structure (first InChIKey block) occurs in the training split. FP oracle : mean over test spectra of the highest Tanimoto similarity between the true fingerprint and any training fingerprint, i.e. the ceiling of any method that returns a single training fingerprint.

		Fallback (+ sub-formula)				Sub-formula rule only
Dataset	Split	$w$	$\alpha$	$T$	$k$	$w$	$\alpha$	$T$	$k$
NPLIB1	Scaffold	4	1	0.05	10	4	1	0.05	10
NPLIB1	Random	4	1	0.1	5	2	4	0.05	10
MassSpecGym	Scaffold	8	1	0.05	20	8	1	0.05	20
MassSpecGym	Random	4	2	0.05	10	4	4	0.01	5

Appendix

Table 4: Hyperparameters of the sub-formula rule selected on the validation split: on the validation spectra without a same-formula training spectrum when the rule is the fallback, and on all validation spectra when it is applied to every test spectrum.

k-Nearest Neighbors

k-Nearest Neighbors (k-NN) predicts or retrieves from nearby data points using local similarity for classification, regression, clustering, and uncertainty.

See all 32 papers

Small molecule retrieval from tandem mass spectrometry: what are we optimizing for?
Feb 18, 2026Gaetan De Waele, Marek Wydmuch, Krzysztof Dembczyński +2Mass Spectrometry
Department of Data Analysis and Mathematical Modelling, Ghent University, Belgium · Snowflake AI Research · Poznan University of Technology
When should we trust the annotation? Selective prediction for molecular structure retrieval from mass spectra
Mar 11, 2026Mira Jürgens, Gaetan De Waele, Morteza Rakhshaninejad +1Selective Prediction Mass Spectrometry
Department of Data Analysis and Mathematical Modeling, Ghent University, Coupure Links 653, Ghent, 9000, Belgium. · Department of Computer Science, University of Antwerp, Middelheimlaan 1, Antwerp, 2020, Belgium.
Hyper-Dimensional Fingerprints as Molecular Representations
Apr 30, 2026Jonas Teufel, Luca Torresi, André Eberhard +1Hyperdimensional Computing Molecular Property Prediction
Karlsruhe Institute of Technology (KIT), Institute of Nanotechnology (INT), Kaiserstraße 12, Karlsruhe, Germany. · Karlsruhe Institute of Technology (KIT), Institute of Anthropomatics and Robotics (IAR), Kaiserstraße 12, Karlsruhe, Germany.

Feb 18, 2026J/K move · Enter open · S save

Small molecule retrieval from tandem mass spectrometry: what are we optimizing for?

Gaetan De Waele, Marek Wydmuch, Krzysztof Dembczyński, Wojciech Kotłowski+1

Department of Data Analysis and Mathematical Modelling, Ghent University, Belgium · Snowflake AI Research · Poznan University of Technology

Mass Spectrometry

Open paper AskPDF arXiv

One of the central challenges in the computational analysis of liquid chromatography-tandem mass spectrometry (LC-MS/MS) data is to identify the compounds underlying the output spectra. In recent years, this problem is increasingly tackled using deep learning methods. A common strategy involves predicting a molecular fingerprint vector from an input mass spectrum, which is then used to search for matches in a chemical compound database. While various loss functions are employed in training these predictive models, their impact on model performance remains poorly understood. In this study, we investigate commonly used loss functions, deriving novel regret bounds that characterize when Bayes-optimal decisions for these objectives must diverge. Our results reveal a fundamental trade-off between the two objectives of (1) fingerprint similarity and (2) molecular retrieval. Optimizing for more accurate fingerprint predictions typically worsens retrieval results, and vice versa. Our theoretical analysis shows this trade-off depends on the similarity structure of candidate sets, providing guidance for loss function and fingerprint selection.

Abstract

k-Nearest Neighbors

Explore similar work

Small molecule retrieval from tandem mass spectrometry: what are we optimizing for?

When should we trust the annotation? Selective prediction for molecular structure retrieval from mass spectra

Hyper-Dimensional Fingerprints as Molecular Representations