Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 43 of the 43 references it cites were found in a registry or an index; none of the 14 sources its claims cite is checked for support yet.

AI-designed NMR spectroscopy RF pulses for fast acquisition at high and ultra-high magnetic fields

Open access
Manu, V. S., Olivieri, Cristina, Veglia, Gianluigi · 2023 · Nature Communications

Abstract

Nuclear magnetic resonance (NMR) spectroscopy is a powerful high-resolution tool for characterizing biomacromolecular structure, dynamics, and interactions. However, the lengthy longitudinal relaxation of the nuclear spins significantly extends the total experimental time, especially at high and ultra-high magnetic field strengths. Although longitudinal relaxation-enhanced techniques have sped up data acquisition, their application has been limited by the chemical shift dispersion. Here we combined an evolutionary algorithm and artificial intelligence to design 1 H and 15 N radio frequency (RF) pulses with variable phase and amplitude that cover significantly broader bandwidths and allow for rapid data acquisition. We re-engineered the basic transverse relaxation optimized spectroscopy experiment and showed that the RF shapes enhance the spectral sensitivity of well-folded proteins up to 180 kDa molecular weight. These RF shapes can be tailored to re-design triple-resonance experiments for accelerating NMR spectroscopy of biomacromolecules at high fields.

Overview

CoChat Summary

TL;DR

Manu et al. use GENETICS-AI to design phase- and amplitude-modulated 1H and 15N RF pulses and incorporate them into a RAPID-TROSY experiment. The new sequence broadens operational bandwidth and improves sensitivity for proteins up to 180 kDa at high magnetic fields, especially at longer interscan delays.

Key findings

  • GENETICS-AI generated four universal spin operations: UA90ev1 (band-selective 90° excitation with chemical shift and J evolution), URev1 (1H refocusing with J evolution), URev2 (15N refocusing with J evolution), and UARev1 (band-selective 180° centered on amides, avoiding water). UA90ev1 selectively excites 0.5–7.5 kHz while leaving water and other protons resonating between 0 and −6.2 kHz unaffected; UARev1 refocuses 0.5–7.5 kHz and avoids water irradiation.
  • Re-engineering the [1H,15N] TROSY-HSQC into RAPID-TROSY improved sensitivity. For RKIP (21 kDa, 187 amide peaks), BEST-TROSY and RAPID-TROSY were most sensitive at D1 0.2–0.25 s; RAPID-TROSY became the most sensitive for D1 > 0.3 s, and at D1 > 1.0 s WADE- and RAPID-TROSY outperformed the other two sequences. For MBP (42 kDa, 279 amide peaks; 139 low-intensity peaks) and RIIβ dimer (180 kDa; 244 resolved peaks in buildup, 122 lower-half peaks, and 390 amide resonances in gain distribution), RAPID-TROSY showed the highest average sensitivity over a wide range of interscan delays.
  • Simulations with 1 ms pulses and R1/R2 from 0 to 50 s−1 showed performance similar to or better than PC9_90 and REBURP, with minimal relaxation effects; URev1 and URev2 had broader operational bandwidth and more homogeneous amide irradiation than REBURP.
  • RAPID-TROSY also showed improved suppression of side-chain amide resonances due to phase cycling and high-fidelity pulse operations.

Methods

  • Pulse design and simulations: GENETICS-AI combined an evolutionary algorithm and AI; simulations used Matlab R2022a, single-spin and two-spin systems (J = −92 Hz), offsets from −10 to +10 kHz, and Bloch-equation relaxation with R1/R2 from 0 to 50 s−1.
  • Protein samples: U-15N labeled RKIP (21 kDa, 0.3 mM), U-13C,15N labeled MBP (42 kDa, 1 mM), and U-15N labeled RIIβ dimer (180 kDa, 0.25 mM), expressed and purified from E. coli as described.
  • NMR spectroscopy: 900 MHz Bruker Avance III with TCI cryoprobe at 300 K; interscan delays varied from 0.2 to 3 s; 2D TROSY-HSQC experiments acquired with 1024 complex points in the direct dimension and 96/160/128 increments for RKIP/MBP/RIIβ, with 16/4/32 scans respectively; processed with NMRPipe and analyzed with NMRFAM-SPARKY.

Limitations

  • The paper does not state a dedicated limitations section or explicit limitations.
  • Evident from the text: only three proteins were tested (21, 42, and 180 kDa) and all NMR experiments were performed at 900 MHz; RAPID-TROSY was demonstrated for [1H,15N] TROSY-HSQC, while implementation in triple-resonance experiments is proposed but not shown.
  • The GENETICS-AI pulses are longer than typical hard pulses; relaxation effects were assessed by simulation with a 1 ms pulse duration, but not exhaustively across all experimental conditions. The RIIβ peak counts also vary across analyses as reported (244 resolved peaks vs. 390 amide resonances in the gain distribution).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41467-023-39581-4

Cites

43 of 43 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 RF pulse shapes designed using the GENETICS-AI software. RF pulse amplitudes and shapes (left), offset responses (center), and J coupling evolution profiles (right) for A. the band-selective 90o pulse UA90ev1, B. the broadband 180o pulse URev1, C. the broadband 180o pulse URev2, and D. the band-selective 180o UARev1. The effects of the pulses on the magnetization components are color-coded in the graph with the offset response. The oscillations of the Mx and My components of the magnetization indicate the chemical shift evolution during the pulse. The J coupling evolution was calculated using the two spin operators, 2IxSz and 2IySz, during the pulse duration, with the UA90ev1 pulse starting from an initial magnetization ρinitial: Iz, the URev1 from ρinitial: Ix, the URev2 from ρinitial: Ix, and the UARev1 pulse from ρinitial: Ix. Note that pulsing on a single channel refocuses the J coupling evolution, whereas pulsing on both channels restores the J coupling evolution. The green traces indicate the offset response for ideal instantaneous pulses. The blue traces are the offset response for pulsing on a single channel (I or S), and the red traces are the offset response for pulsing on both channels (I and S). The duration and operational bandwidth of the pulses are reported in Supplementary Table 1, and the 2D response profiles are shown in Supplementary Fig. 1.
Fig. 1 RF pulse shapes designed using the GENETICS-AI software. RF pulse amplitudes and shapes (left), offset responses (center), and J coupling evolution profiles (right) for A. the band-selective 90o pulse UA90ev1, B. the broadband 180o pulse URev1, C. the broadband 180o pulse URev2, and D. the band-selective 180o UARev1. The effects of the pulses on the magnetization components are color-coded in the graph with the offset response. The oscillations of the Mx and My components of the magnetization indicate the chemical shift evolution during the pulse. The J coupling evolution was calculated using the two spin operators, 2IxSz and 2IySz, during the pulse duration, with the UA90ev1 pulse starting from an initial magnetization ρinitial: Iz, the URev1 from ρinitial: Ix, the URev2 from ρinitial: Ix, and the UARev1 pulse from ρinitial: Ix. Note that pulsing on a single channel refocuses the J coupling evolution, whereas pulsing on both channels restores the J coupling evolution. The green traces indicate the offset response for ideal instantaneous pulses. The blue traces are the offset response for pulsing on a single channel (I or S), and the red traces are the offset response for pulsing on both channels (I and S). The duration and operational bandwidth of the pulses are reported in Supplementary Table 1, and the 2D response profiles are shown in Supplementary Fig. 1.CC BY · PubMed Central
Fig. 2 Pulse scheme for the [1H,15N] RAPID-TROSY experiment. The pulses generated by the GENETICS-AI software for 1H and 15N channels are reported according to the optimized amplitude shapes generated. Open rectangular pulses in the 15N channel are 90° hard pulses and were not substituted in the original sequence. The pulse operations for the GENETICS-AI pulses are summarized in Table S1 and explained in the main text. The delay, Δ, associated with UARev1 is 0.75 times the length of the UARev1 pulse. The delay τ is equal to 1/(2JHN), where JHN is the coupling constant between 1H and 15N. The phase cycle φ1 = (y, −y, −x, x) for odd and φ1 = (−y, y, −x, x) for even t1 points. Receiver phase cycle φr = (x, −x, −y, y). The 1H magnetization of water (W) and aliphatic protons is kept along the z-axis throughout the pulse sequence. The amplitude of the gradient pulses, g1, g2, and g3, applied along the z-axis were 1.97, 2.96, and 3.29 Gauss.
Fig. 2 Pulse scheme for the [1H,15N] RAPID-TROSY experiment. The pulses generated by the GENETICS-AI software for 1H and 15N channels are reported according to the optimized amplitude shapes generated. Open rectangular pulses in the 15N channel are 90° hard pulses and were not substituted in the original sequence. The pulse operations for the GENETICS-AI pulses are summarized in Table S1 and explained in the main text. The delay, Δ, associated with UARev1 is 0.75 times the length of the UARev1 pulse. The delay τ is equal to 1/(2JHN), where JHN is the coupling constant between 1H and 15N. The phase cycle φ1 = (y, −y, −x, x) for odd and φ1 = (−y, y, −x, x) for even t1 points. Receiver phase cycle φr = (x, −x, −y, y). The 1H magnetization of water (W) and aliphatic protons is kept along the z-axis throughout the pulse sequence. The amplitude of the gradient pulses, g1, g2, and g3, applied along the z-axis were 1.97, 2.96, and 3.29 Gauss.CC BY · PubMed Central
Fig. 3 Sensitivity comparison of the different TROSY schemes for RKIP. A Buildup of the average sensitivity calculated for 187 amide peaks of U-15N labeled RKIP. B Sensitivity comparison of the different TROSY experiments as a function of the interscan delay. The average intensity was divided by the root square of the total scan duration, Tscan, which includes the duration of the pulse sequence, the acquisition time, and the interscan delay. C Sensitivity gain distribution of RAPID-TROSY for all the 187 resolved amide resonances of RKIP compared with the other three pulse sequences at different interscan delays. The sensitivity gain for each resonance is calculated using(IRAPID−I)/I, where IRAPID indicates the intensity of each RAPID-TROSY peak, and I indicates the intensity of the corresponding resonance in the other sequences.
Fig. 3 Sensitivity comparison of the different TROSY schemes for RKIP. A Buildup of the average sensitivity calculated for 187 amide peaks of U-15N labeled RKIP. B Sensitivity comparison of the different TROSY experiments as a function of the interscan delay. The average intensity was divided by the root square of the total scan duration, Tscan, which includes the duration of the pulse sequence, the acquisition time, and the interscan delay. C Sensitivity gain distribution of RAPID-TROSY for all the 187 resolved amide resonances of RKIP compared with the other three pulse sequences at different interscan delays. The sensitivity gain for each resonance is calculated using(IRAPID−I)/I, where IRAPID indicates the intensity of each RAPID-TROSY peak, and I indicates the intensity of the corresponding resonance in the other sequences.CC BY · PubMed Central
Fig. 4 Comparison of the performance of the different TROSY schemes with interscan delays of 0.2 and 2.0 s. A RAPID-TROSY, B BEST-TROSY, and C conventional Bruker TROSY experiment (trosyetf3gpsi2). D, E 1D traces for the resonances 1–6 labeled in the A panel acquired with interscan delays of 0.2 and 2.0 s. The 1D spectra are analyzed at the same noise levels. The red, blue, and green 1D peaks correspond to the RAPID-TROSY, BEST-TROSY, and trosyetf3gpsi2 experiments.
Fig. 4 Comparison of the performance of the different TROSY schemes with interscan delays of 0.2 and 2.0 s. A RAPID-TROSY, B BEST-TROSY, and C conventional Bruker TROSY experiment (trosyetf3gpsi2). D, E 1D traces for the resonances 1–6 labeled in the A panel acquired with interscan delays of 0.2 and 2.0 s. The 1D spectra are analyzed at the same noise levels. The red, blue, and green 1D peaks correspond to the RAPID-TROSY, BEST-TROSY, and trosyetf3gpsi2 experiments.CC BY · PubMed Central
Fig. 5 Comparison of the different TROSY pulse sequences for MBP. A RAPID-TROSY spectrum of U-15N labeled MBP with an interscan delay of 0.25 s. B Average intensity of the 279 amide peaks vs. interscan delay for the four TROSY pulse sequences. C Average intensity of the 139 low-intensity (lower half) peaks vs. interscan delay. D Sensitivity gain distribution of RAPID-TROSY for all resolved amide resonances compared with the other three pulse sequences at different interscan delays. The sensitivity gain for each resonance is calculated using(IRAPID−I)/I, where IRAPID indicates the intensity of each RAPID-TROSY peak, and I indicates the intensity of the corresponding resonance in the other sequences. All spectra at different interscan delays are shown in Supplementary Fig. 6.
Fig. 5 Comparison of the different TROSY pulse sequences for MBP. A RAPID-TROSY spectrum of U-15N labeled MBP with an interscan delay of 0.25 s. B Average intensity of the 279 amide peaks vs. interscan delay for the four TROSY pulse sequences. C Average intensity of the 139 low-intensity (lower half) peaks vs. interscan delay. D Sensitivity gain distribution of RAPID-TROSY for all resolved amide resonances compared with the other three pulse sequences at different interscan delays. The sensitivity gain for each resonance is calculated using(IRAPID−I)/I, where IRAPID indicates the intensity of each RAPID-TROSY peak, and I indicates the intensity of the corresponding resonance in the other sequences. All spectra at different interscan delays are shown in Supplementary Fig. 6.CC BY · PubMed Central
Fig. 6 Comparison of the different TROSY pulse sequences for RIIβ dimer. A RAPID-TROSY of 1H,15N labeled PKA RIIβ dimer acquired with an interscan delay of 0.25 s. B Average intensity of the amide peaks vs. interscan delay for all resolved peaks (244); C Average intensity of the less intense peaks (122, lower half) vs. interscan delay. D Sensitivity gain distribution of RAPID-TROSY for all the resolved 390 amide resonances of RIIβ compared with the other three pulse sequences at different interscan delays. All spectra at different relaxation delays are shown in Supplementary Fig. 7.
Fig. 6 Comparison of the different TROSY pulse sequences for RIIβ dimer. A RAPID-TROSY of 1H,15N labeled PKA RIIβ dimer acquired with an interscan delay of 0.25 s. B Average intensity of the amide peaks vs. interscan delay for all resolved peaks (244); C Average intensity of the less intense peaks (122, lower half) vs. interscan delay. D Sensitivity gain distribution of RAPID-TROSY for all the resolved 390 amide resonances of RIIβ compared with the other three pulse sequences at different interscan delays. All spectra at different relaxation delays are shown in Supplementary Fig. 7.CC BY · PubMed Central

Verification

Claims the paper makes

0 Valid · 3 Not valid · 12 Not checked · 64 Own results and interpretations

41 more claims

Create a free account to see all 79 claims and to verify the 11 references still open.

See all claims — free account
  • It may state two separate claims as one. 29
  • Probably not a substantial claim. 10
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 43 references

Research works printed but not linked: 1

  1. Ernst, R. R., Bodenhausen, G. & Wokaun, A. Principle of Nuclear Magnetic Resonance, (Claredon Press, 1987).

Cited by

See all 35

Metadata

Authors
Manu, V. S., Olivieri, C., & Veglia, G.
Affiliations
University of Minnesota (US)
Corresponding author
Veglia, Gianluigi
Venue
Nature Communications
Published
2023-07-12
Volume
14
Issue
1
Pages
4144-4144
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4384025559
Cited by
35
References
43
Citations by year
2026: 11 · 2025: 15 · 2024: 8 · 2023: 1
FWCI
3.07
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
U.S. Department of Health & Human Services | National Institutes of Health (HL144130, GM121735)
Topics
  • Physical Sciences › Chemistry › Spectroscopy › Advanced NMR Techniques and Applications
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced MRI Techniques and Applications
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
Keywords
RF pulse design, NMR spectroscopy, high magnetic fields, ultrahigh magnetic fields, evolutionary algorithms, artificial intelligence, chemical shift dispersion, spectral sensitivity, biological macromolecules, spin-lattice relaxation
MeSH terms
Algorithms, Artificial Intelligence, Heart Rate, Magnetic Resonance Spectroscopy, Magnetic Fields

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: AI-designed NMR spectroscopy RF pulses for fast acquisition at high and ultra-high magnetic fields. https://app.cochat.ai/papers/10.1038/s41467-023-39581-4
The paper
Manu, V. S., Olivieri, C., & Veglia, G. (2023). AI-designed NMR spectroscopy RF pulses for fast acquisition at high and ultra-high magnetic fields. Nature Communications. https://doi.org/10.1038/s41467-023-39581-4

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.