Units
1. Introduction to Artificial Intelligence in Chemistry: Roles, Boundaries, Validation and Ethics 2. Molecular Representation and Chemical Structure: Validation with SMILES, InChI and RDKit 3. Reaction Prediction and Mechanism: Product, Condition and Adverse Reaction Prediction 4. Spectrum Interpretation Support: NMR, IR and Mass Spectrometry 5. Literature Review and Synthesis Planning: Source, Procedure and Roadmap 6. Chemical Data Analysis and Python: pandas, Stoichiometry and Calibration 7. Molecular Property Prediction and QSAR: Resolution, pKa and Model Limits 8. Laboratory Documentation: Experiment Notebook, ELN and Reproducibility 9. Safety and Hazard Assessment: GHS, SDS and the Limits of Artificial Intelligence 10. Confirmation, Hallucination, and Scientific Integrity 11. Human in Experimentation: Ethics, Privacy, Responsibility and End-to-End Workflow
Unit 4 / 11

Spectrum Interpretation Support: NMR, IR and Mass Spectrometry

Gains:

  • Ability to understand what NMR, IR and MS show and use artificial intelligence as a hypothesis tester based on the expected structure
  • Ability to check assignment consistency by providing complete details such as integration, splitting, solver and ppm threshold
  • Ability to understand that the final structure assignment rests with the chemist examining the experimental spectrum and that multiple techniques must be used together.

When you synthesize a compound, the real question is: "Did I really get the molecule I wanted?" This is answered by spectroscopy. Three basic techniques: NMR (reads the magnetic environment of nuclei and shows the neighborhood of atoms), IR (reads the vibration of bonds in the molecule and shows functional groups), and mass spectrometry (MS) (shows the mass and fragmentation pattern of the molecule). AI is a powerful assistant in interpreting these spectra: groups peaks, suggests possible assignments, checks consistency. But the final structure assignment is made by the chemist who sees and evaluates the experimental spectrum. In this unit, we will learn to use AI in spectrum interpretation in a supervised manner.

What the three techniques say

  • 1H NMR: Shows the number of hydrogen atoms, their neighbors (cleavage pattern), and chemical environment (shift value, ppm). For example, at ~7 ppm the signals are suggestive of aromatic hydrogens, at ~2 ppm a singlet acetyl group.
  • 13C NMR: Shows the carbon skeleton; the carbonyl carbon appears at ~170-200 ppm.
  • IR: Indicates functional groups; ~1700 cm⁻¹ carbonyl (C=O), ~3300 cm⁻¹ broadband O-H or N-H.
  • MS: Molecular ion peak (M⁺) gives the molecular weight, fragmentations give structure clues. High resolution MS (HRMS) confirms the molecular formula.

AI knows the typical ranges of these values ​​well. So "What could 1710 cm⁻¹ be?" It answers the question reliably. Where it is weak is in making assignments without seeing the fine detail of the actual experimental spectrum (overlapping peaks, impurity signals).

Hint: Give the spectrum to the AI ​​together as “raw data + expected structure” and ask “is this data consistent with this structure?” ask. Using AI as a hypothesis tester asks “what compound is this spectrum?” It is much more reliable than asking from scratch.

Step by step: Spectrum interpretation with AI

  1. Give the expected structure: Give the SMILES of the compound you synthesized. Let the AI ​​tell you what to expect first.
  2. Export raw data: Enter peak lists (NMR shifts, integrations, splits; IR bands; MS m/z values) in an orderly manner.
  3. Ask a consistency question: "Are these data consistent with the expected structure? Which peak corresponds to which atom?"
  4. Look for gaps: "Is there a peak that is expected but not visible? Is there an unexpected peak (impurity)?"
  5. Alternative test: If there is inconsistency, "what other structure explains this data better?"
  6. Human decision: You confirm the final assignment by examining the experimental spectrum.

Four copyable templates

1) 1H NMR consistency check:

Expected compound: 4-methoxyacetophenone (SMILES: COc1ccc(cc1)C(C)=O).1H NMR (CDCl3) peak list:- 7.93 ppm (2H, d)- 6.93 ppm (2H, d)- 3.87 ppm (3H, s)- 2.56 ppm (3H, s)Task:1) Each peak is assigned to a hydrogen group ata.2) Are the total H number and integration consistent?3) Are there any missing/unexpected peaks?The final decision is up to me; Give your consistency analysis.

2) IR functional group:

IR (cm^-1) significant bands: 1675, 1600, 1258, 2840. The expected compound contains an aromatic ketone and a methoxy group. Task: Assign each band to a possible bond/group. Explain why 1675 cm^-1 and not 1715 (conjugation effect?).

3) MS formula verification:

[M+H]+ = 151.0754 measured by HRMS. Expected compound: 4-methoxyacetophenone (C9H10O2). Task: Calculate the theoretical mass of [M+H]+ for C9H10O2, find the difference in ppm with the measured one, comment if it is consistent. Note: I will also check the exact monoisotopic masses.

4) Multi-technique combining:

The following three data belong to a compound:- IR: wide 3300 cm^-1, 1710 cm^-1.- 1H NMR: 12 ppm wide 1H, 2.4 ppm 2H t, ...- MS: M+ = 88. Task: Evaluate these three data together and suggest a possible structure. Write separately which structural feature each evidence supports. List the alternative structures and explain why you eliminated them.

Weak prompt / Strong prompt

Weak:

What is this NMR? There are peaks at 7.9 and 3.9 ppm.

No context (solvent, integration, cleavage, expected structure). AI randomly predicts a compound.

Strong:

Expected compound: anisole derivative. 1H NMR (CDCl3): 7.93 (2H, d), 6.93 (2H, d), 3.87 (3H, s), 2.56 (3H, s). Task: Check the consistency of the peaks with this structure, discard each peak, suggest an alternative structure if there is inconsistency.

The difference: expected structure, full peak list (integration + cleavage + solvent) and a clear consistency task. AI now tests hypotheses, not makes predictions.

Comparison of techniques

technical

What shows

Is AI support strong?

verification point

1H NMR

H number, neighborhood, environment

Strong (assignment, integration)

Integration sum = number of molecules H

13C NMR

carbon skeleton

Powerful (slip intervals)

Number of carbons must match

IR

Functional groups

Strong (band assignment)

Carbonyl, OH/NH bands

MS/HRMS

Mass, formula

Medium-strong (formula calculation)

Theoretical mass in ppm

mini cases

Case 1 — Incomplete peak capture. A student checked the 1H NMR of his product with AI. YZ noticed that an OH peak that was supposed to be in the expected structure was not on the list and said "it could be a proton that is exchangeable or the product is wrong." Student did D2O shake test, OH was indeed present but overlapped. AI's "missing peak" warning caused the right question to be asked.

Case 2 — Incorrect assignment correction. YZ initially assigned a singlet at 2.1 ppm to methyl ketone; but the integration was 2H instead of 3H. The user stated this, the AI ​​updated the assignment to a CH2 group. Lesson: make the integration clear to the AI, or it will throw it wrong.

Case 3 — HRMS ppm error. A compound C10H12O2 was expected ([M+H]+ theoretical 165.0910). Measured 165.0912; difference 1.2 ppm is acceptable (<5 ppm). The AI ​​did this calculation, but the user also checked the monoisotopic masses with an independent tool; values ​​overlapped. Lesson: Do MS formula validation with ppm threshold, don't say "close" by eye.

Common mistakes

  • Asking without giving the expected structure. Forcing AI to guess from scratch increases the error rate; Test the hypothesis.
  • Skipping integration/cleavage. Without this information in NMR, assignment is unreliable.
  • Forgetting the solvent peak. Assigning solvent signals such as CDCl3 ~7.26 ppm, DMSO ~2.50 ppm to the compound.
  • Ignoring the isotope pattern in MS. In compounds containing chlorine/bromine, the M+2 pattern gives information about the structure.
  • Relying on a single technique. Considering NMR, IR, and MS together is much more robust than relying on either alone.
  • To approve the experimental spectrum without seeing it. AI interprets the peak list; One must look at the actual spectrum for details such as overlap, impurities, etc.
Attention: Just because the AI ​​says "consistent" in the spectrum interpretation does not mean that the structure is absolutely correct; it just means "this data does not rule out this structure". Accurate assignment requires multiple techniques and human judgment.

In summary

  • NMR indicates neighborhood, IR indicates functional group, MS indicates mass/formula; They are powerful when used together.
  • Use AI as a hypothesis tester: give expected structure + raw data and ask for consistency.
  • Enter complete details such as integration, splitting, solvent and ppm threshold.
  • Perform HRMS formula validation with numerical ppm threshold; Do not neglect the MS isotope pattern.
  • The final structure assignment is up to the chemist examining the experimental spectrum.

Application task

Get actual NMR, IR, and MS data of a compound you synthesized (or retrieved from a literature article). Give the AI ​​SMILES first and ask "what should I expect?" ', then give the raw data and have it checked for consistency. Try to find one correct and one incorrect/incomplete assignment of the AI ​​in each technique. Calculate the theoretical and measured mass difference in ppm for HRMS. Plot the findings into a table.

checklist

  • [ ] I know what NMR, IR and MS show.
  • [ ] I give the expected structure to the AI ​​and test the hypothesis.
  • [ ] I enter the integration, splitting, solvent and peak list completely.
  • [ ] I make HRMS verification numerical with ppm threshold.
  • [ ] I evaluate multiple techniques together.
  • [ ] I confirm the final assignment by examining the experimental spectrum.
Previous Next