Skip to main content
Tessa NewsLink
Paste a health news link, or browse

Source study found

Story checked

New AI tool predicts hydrogen positions to speed up drug discovery (opens in a new tab)

news-medical.net · 2026-09-23

Short answerEvidenceSource

Short answer

Mixed

Mixed.

The claims we could check match the study, but some claims were not covered by the evidence reviewed.

  • 2 supported
  • 3 not covered

Checked against the study summary. The full text wasn't available, so some details couldn't be settled either way.

Share this check

Follow the evidence trail
1
2

NewsLink checks it

Mixed

Every claim we could check holds up. Two of five claims match the study. This overall rating is based only on the claims we could check. Three claims the study doesn't address.

  • 2 supported
  • 3 not covered
Open claim evidence
3
Source paper

Source layer

The 2 papers the story cites

Source study separated from background citations.

The research anchor for the report.

  • The study this story reportspresented as the new finding

    Deep learning of tautomer stability from crystallographic proton positions

    Chemical Science · 2026

    Publisher (opens in a new tab)
  • The study this story reportspresented as the new finding

    Deep learning of tautomer stability from crystallographic proton positions

    Publisher (opens in a new tab)
Then inspect each claim

Evidence layer

Claim by claim

Each claim gets a verdict. Expand it to see the evidence directly below.

5 claims in this story

Showing all 5 claimsChoose a verdict to focus the list.

Claim 1 of 5Not coveredThe team mined the Cambridge Structural Database to build a dataset of more than 1.1 million tautomeric states and trained a graph neural network to predict stable tautomers directly from 2D molecular forms without requiring 3D structures or quantum-mechanical calculations.View evidenceHide evidence

As statedmore than 1.1 million tautomeric states

Why this verdict

The abstract-level profile supports that crystallographic proton positions were mined to curate tautomer labels and train a GNN that predicts stable tautomers from 2D topology. However, the abstract profile does not verify the Cambridge Structural Database specifically, the stated size of more than 1.1 million tautomeric states, or detailed claims about not requiring 3D structures or quantum-mechanical calculations. The core method is consistent, but the quantitative and source-specific details are not verifiable at abstract depth.

1

Study evidence

Crystallographic proton positions were used to create a curated dataset of stable tautomer labels that served as ground truth for training and validating a graph neural network to predict tautomeric states from 2D topology.

“By mining crystallographic proton positions”
2

Study evidence

A graph neural network trained on tautomer labels derived from crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology, enabling rapid tautomer assignment.

“we trained a graph neural network to predict stable tautomeric states directly from 2D molecular topology”
Claim 2 of 5Not coveredWhen applied to 5,075 PDBbind ligands with multiple possible tautomeric states, the model identified 126 cases, about 2.5 percent, in which the assigned ligand tautomer was likely incorrect.View evidenceHide evidence

As stated126 cases, approximately 2.5 percent

Why this verdict

The abstract-level profile contains only a general claim that the method enables rapid tautomer assignment and lacks downstream application details. It does not report evaluation on 5,075 PDBbind ligands, identification of 126 likely incorrect tautomer assignments, or the approximately 2.5% figure. These may be in the full paper, but they are not verifiable from the supplied abstract-level evidence.

1

Study evidence

A graph neural network trained on crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology.

“enabling rapid tautomer assignment for high-throughput molecular discovery”
Claim 3 of 5Not coveredThe researchers released the method as an open-source tool, Tautomer-Predictor, which the article says can rapidly analyze very large molecular libraries and processed about 4.6 million compounds in 3.2 hours on a single GPU-enabled node in one test.View evidenceHide evidence

As statedabout 4.6 million compounds in 3.2 hours

Why this verdict

The abstract-level profile supports only a qualitative claim that the method enables rapid tautomer assignment for high-throughput discovery. It does not verify release as an open-source tool named Tautomer-Predictor, nor the runtime benchmark of about 4.6 million compounds in 3.2 hours on a single GPU-enabled node. The headline utility is directionally consistent, but the tool-release and quantitative throughput details are not verifiable at this depth.

1

Study evidence

A graph neural network trained on crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology.

“enabling rapid tautomer assignment for high-throughput molecular discovery”
Claim 4 of 5SupportedNew York University researchers trained an AI model to learn chemical patterns associated with stability in drug-like molecules and predict where their hydrogen atoms should be positioned.View evidenceHide evidence

Why this verdict

The abstract-level profile supports the core claim that the authors mined crystallographic proton positions and trained a graph neural network to predict stable tautomeric states from 2D molecular topology. Framing this as predicting hydrogen/proton placement is consistent with tautomer assignment from proton positions, although details such as the researchers' NYU affiliation and the exact scope 'drug-like molecules' are not independently detailed in the supplied abstract profile.

1

Study evidence

Crystallographic proton positions were used to create a curated dataset of stable tautomer labels that served as ground truth for training and validating a graph neural network to predict tautomeric states from 2D topology.

“By mining crystallographic proton positions”
2

Study evidence

A graph neural network trained on tautomer labels derived from crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology, enabling rapid tautomer assignment.

“we trained a graph neural network to predict stable tautomeric states directly from 2D molecular topology”
Claim 5 of 5SupportedThe research was published in Chemical Science and addresses the challenge of rapidly determining the stable form of molecules that can convert into different tautomeric forms.View evidenceHide evidence

Why this verdict

The supplied profile supports that the work addresses rapid assignment of stable tautomeric states for molecules capable of tautomerism and frames this as useful for high-throughput molecular discovery. The journal venue is not a scientific result and is not described in the unit evidence, but the profile is for the supplied paper record; the substantive research claim is supported at abstract depth.

1

Study evidence

A graph neural network trained on tautomer labels derived from crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology, enabling rapid tautomer assignment.

“we trained a graph neural network to predict stable tautomeric states directly from 2D molecular topology”
2

Study evidence

A graph neural network trained on crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology.

“enabling rapid tautomer assignment for high-throughput molecular discovery”
Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

  • The profile notes potential repository and applicability-domain limits, including possible biases in crystallographic databases and uncertain generalisability beyond represented chemical space.

    The story emphasizes broad usefulness for large-scale drug-discovery workflows but does not mention the abstract-level limitations about representativeness of crystallographic training data or generalisability to diverse molecular libraries.

    From secondary_data; Graph neural network supervised learning; in silico

3 things the story did carry across
  • The paper's primary methodological contribution is training a graph neural network on crystallographic proton-position-derived labels to predict stable tautomeric states from 2D molecular topology.
  • The paper includes a dataset-curation component: mining deposited crystallographic structures/proton positions to derive ground-truth stable-tautomer labels for training and validation.
  • The paper presents the model as enabling rapid tautomer assignment for high-throughput molecular discovery workflows.
Then read the study layer

Study layer

Study at a glance

Scan the study first. Expand only the parts you want to inspect.

Pieces of work

3

Evidence read

study summary

Lead result

in silico

1Lead resultin silicoTrain a graph neural network (GNN) using crystallographic proton positions to predict the most stable tautomeric state directly from 2D molecular topology for rapid tautomer assignment in high-throughput discovery.Graph neural network supervised learningExpand

In plain English

The paper trains a supervised graph neural network (GNN) using labels derived from crystallographic proton positions to predict the most stable tautomeric state from 2D molecular topology, with the stated goal of enabling rapid tautomer assignment for high-throughput molecular discovery.

Key findings

  • A graph neural network trained on tautomer labels derived from crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology, enabling rapid tautomer assignment.
“we trained a graph neural network to predict stable tautomeric states directly from 2D molecular topology”
What this piece can’t prove
  • Abstract does not provide quantitative performance metrics, dataset size, or dataset composition.

2 further details could not be confirmed from the summary.

2secondary dataTrain a graph neural network (GNN) using crystallographic proton positions to predict the most stable tautomeric state directly from 2D molecular topology for rapid tautomer assignment in high-throughput discovery.secondary dataExpand

In plain English

The study mined crystallographic proton positions from deposited crystal structures to derive ground-truth labels for the most stable tautomeric state of molecules, assembled a curated dataset from these labels, and used that dataset for training/validation of a graph neural network to predict stable tautomers directly from 2D molecular topology, enabling rapid tautomer assignment for high-throughput discovery.

Key findings

  • Crystallographic proton positions were used to create a curated dataset of stable tautomer labels that served as ground truth for training and validating a graph neural network to predict tautomeric states from 2D topology.
“By mining crystallographic proton positions”
What this piece can’t prove
  • Abstract does not report procedures to prevent leakage between training/validation/test splits or quantify label noise.

3 further details could not be confirmed from the summary.

3in silicoTrain a graph neural network (GNN) using crystallographic proton positions to predict the most stable tautomeric state directly from 2D molecular topology for rapid tautomer assignment in high-throughput discovery.Expand

In plain English

The paper reports training a graph neural network on crystallographic proton positions to predict the most stable tautomeric state from 2D molecular topology, and states that this enables rapid tautomer assignment for high-throughput molecular discovery.

Key findings

  • A graph neural network trained on crystallographic proton positions can predict the most stable tautomeric state from 2D molecular topology.
  • The method is presented as enabling rapid tautomer assignment for high-throughput molecular discovery.
“enabling rapid tautomer assignment for high-throughput molecular discovery”
What this piece can’t prove
  • Abstract does not report error analysis, failure modes, or applicability domain for the model.

2 further details could not be confirmed from the summary.

Finally, the search trail

Method layer

NewsLink found the paper. Tessa takes you deeper.

NewsLink checks the story. Tessa is where you inspect the paper, authors, evidence, and research context.

Open the paper in Tessa

Deep learning of tautomer stability from crystallographic proton positions

Chemical Science · 2026

Source record (opens in a new tab)

Why this one

Near certain

NewsLink found the paper. Tessa is where you inspect it deeply.

Papers considered

The selected paper, plus nearby candidates.

Crossref, PubMed · 41 candidate papers

Candidate

X-ray charged afterglow nanoplatform for cascade-targeted, oxygen-independent eradication of intracellular bacteria

Chemical Engineering Journal · 2026 · Crossref

Source (opens in a new tab)
Candidate

A hermetically-isolated magnetic coupling triboelectric nanogenerator enabled self-powered in-line milk sterilization

Chemical Engineering Journal · 2026 · Crossref

Source (opens in a new tab)

And 35 more candidates considered.