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

Source study found

Story checked

Can AI find drugs to fight infection? (opens in a new tab)

news-medical.net · 2026-09-17

Short answerEvidenceSource

Short answer

Mostly supported

Mostly supported.

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

  • 4 supported
  • 1 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

Mostly supported

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

  • 4 supported
  • 1 not covered
Open claim evidence
3
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 senior author said AI-guided drug repurposing is a powerful strategy for combating antimicrobial resistance, especially for pathogens with some known active molecules that can be used for training or fine-tuning.View evidenceHide evidence

Why this verdict

The profile supports the general idea that AI-guided repurposing was used and that known active/inactive compounds formed the training data, but the supplied abstract-level profile does not verify the senior-author attribution or the quoted characterization that the strategy 'has become a powerful strategy' for antimicrobial resistance. The quote and its attribution require evidence beyond the abstract-level profile.

1

Study evidence

Ensembles of transformer, graph, and tree models were trained on 1,849 actives and 34,503 inactives and used to prospectively score 6,747 drugs for activity against S. pneumoniae.

“we leveraged ensembles of transformer, graph, and tree models, each trained on a set of 1849 actives along with 34 503 inactives, to prospectively examine 6747 drugs.”
Claim 2 of 5SupportedResearchers used artificial intelligence to identify approved drugs that may be effective against Streptococcus pneumoniae.View evidenceHide evidence

Why this verdict

The abstract-level profile supports that the researchers used AI/ML ensembles to prospectively prioritize repurposable drugs against Streptococcus pneumoniae and then validated selected candidates in vitro. The story's 'may be effective' wording is appropriately hedged, provided it is understood as candidate/in vitro activity rather than proven clinical efficacy.

1

Study evidence

Ensembles of transformer, graph, and tree models were trained on 1,849 actives and 34,503 inactives and used to prospectively score 6,747 drugs for activity against S. pneumoniae.

“we leveraged ensembles of transformer, graph, and tree models, each trained on a set of 1849 actives along with 34 503 inactives, to prospectively examine 6747 drugs.”
2

Study evidence

Screen of 11 AI-prioritized candidate antibiotics against S. pneumoniae R6 found nine compounds that strongly reduced in vitro growth (IC50 ≤ 0.4 µg/mL).IC50 ≤ 0.4 µg/mL (for nine validated compounds)

“Of 11 selected candidate antibiotics, nine were found to strongly reduce in vitro growth of S. pneumoniae R6, with IC50 values of ≤ 0.4 µg/mL.”
Claim 3 of 5SupportedThe team used three learning approaches—ensembles of decision trees, graph neural networks, and sequence-based transformers—to screen almost 7,000 candidate drugs for inhibition of S. pneumoniae.View evidenceHide evidence

As statedalmost 7,000 candidate drugs

Why this verdict

The profile states that ensembles of transformer, graph, and tree models were trained on labeled antibacterial activity data and used to examine 6,747 drugs. That supports the story's description of three learning approaches and 'almost 7,000' candidate drugs.

1

Study evidence

Ensembles of transformer, graph, and tree models were trained on 1,849 actives and 34,503 inactives and used to prospectively score 6,747 drugs for activity against S. pneumoniae.

“we leveraged ensembles of transformer, graph, and tree models, each trained on a set of 1849 actives along with 34 503 inactives, to prospectively examine 6747 drugs.”
Claim 4 of 5SupportedOf the nearly 7,000 candidates, 11 were selected for experimental validation and 9 inhibited the growth of S. pneumoniae.View evidenceHide evidence

As stated11 selected; 9 active

Why this verdict

The abstract-level evidence directly reports that 11 selected candidate antibiotics were tested and that nine strongly reduced in vitro growth of S. pneumoniae R6, matching the stated numbers.

1

Study evidence

Ensembles of transformer, graph, and tree models were trained on 1,849 actives and 34,503 inactives and used to prospectively score 6,747 drugs for activity against S. pneumoniae.

“we leveraged ensembles of transformer, graph, and tree models, each trained on a set of 1849 actives along with 34 503 inactives, to prospectively examine 6747 drugs.”
2

Study evidence

Screen of 11 AI-prioritized candidate antibiotics against S. pneumoniae R6 found nine compounds that strongly reduced in vitro growth (IC50 ≤ 0.4 µg/mL).IC50 ≤ 0.4 µg/mL (for nine validated compounds)

“Of 11 selected candidate antibiotics, nine were found to strongly reduce in vitro growth of S. pneumoniae R6, with IC50 values of ≤ 0.4 µg/mL.”
Claim 5 of 5SupportedOne of the two most potent repurposed drugs was effective even against drug-resistant strains of S. pneumoniae.View evidenceHide evidence

As statedone of the two most potent repurposed drugs

Why this verdict

The profile identifies thiostrepton and ceftiofur as the two most potent hits and states that thiostrepton remained highly potent against multidrug-resistant S. pneumoniae strains. This supports the claim, though only at the level of in vitro activity; the abstract does not provide strain counts, identities, or clinical efficacy data.

1

Study evidence

Screen of 11 AI-prioritized candidate antibiotics against S. pneumoniae R6 found nine compounds that strongly reduced in vitro growth (IC50 ≤ 0.4 µg/mL).IC50 ≤ 0.4 µg/mL (for nine validated compounds)

“Of 11 selected candidate antibiotics, nine were found to strongly reduce in vitro growth of S. pneumoniae R6, with IC50 values of ≤ 0.4 µg/mL.”
2

Study evidence

Thiostrepton retained high in vitro potency against multidrug-resistant Streptococcus pneumoniae strains as reported in the abstract.

“Thiostrepton remained highly potent even against multidrug-resistant strains”
Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

  • The models were trained on 1,849 active and 34,503 inactive compounds.

    The story mentions model approaches and candidate screening but does not report the training dataset composition or size, which is material to understanding how the AI models were built.

    From in silico

  • Thiostrepton and ceftiofur were the most potent hits, with abstract-reported IC50 values of 0.0001 µg/mL and 0.0004 µg/mL, respectively.

    The story refers generically to one of the two most potent drugs but, as presented, does not name the two top hits or provide their potency values.

    From in vitro growth inhibition (dose–response) assay

  • The abstract does not specify the number, identities, or resistance profiles of the multidrug-resistant strains, nor quantitative potency metrics for those strains.

    The story notes activity against resistant strains but does not convey the abstract-level uncertainty about the MDR strain panel or the lack of reported quantitative MDR potency data.

    From in vitro susceptibility testing across MDR strain panel

  • The abstract does not report model validation procedures, quantitative predictive performance, candidate-selection thresholds, assay replicates, or statistical uncertainty.

    These abstract-level methodological limitations are not mentioned in the story presentation. They do not negate the reported findings, but they limit how strongly the AI model performance and experimental precision can be assessed from the abstract alone.

    From in silico; in vitro growth inhibition (dose–response) assay

4 things the story did carry across
  • The paper's central method was AI-guided phenotypic drug repurposing using transformer, graph, and tree-model ensembles trained on labeled antibacterial activity data and applied to 6,747 drugs.
  • Eleven AI-prioritized candidates were selected for in vitro validation, and nine strongly inhibited S. pneumoniae R6 growth with IC50 values ≤ 0.4 µg/mL.
  • Thiostrepton retained high in vitro potency against multidrug-resistant S. pneumoniae strains.
  • The evidence is in silico prioritization plus in vitro growth-inhibition testing; it does not establish in vivo efficacy, clinical effectiveness, pharmacokinetics, toxicity, or safety.
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 silicoDevelop and apply AI/ML models (transformer, graph, tree ensembles) trained on labeled antibacterial activity data to prospectively prioritize repurposable drugs against Streptococcus pneumoniae.Expand

In plain English

Ensembles combining transformer, graph, and tree-based supervised models were trained on a labeled antibacterial activity set (1,849 actives and 34,503 inactives) and used to prospectively score 6,747 drugs for phenotypic activity against Streptococcus pneumoniae; 11 top candidates were selected for experimental testing, and nine showed strong in vitro growth inhibition (IC50 ≤ 0.4 µg/mL), including thiostrepton and ceftiofur with sub-nanogram-per-milliliter potencies. Thiostrepton remained highly potent against multidrug-resistant strains.

Key findings

  • Ensembles of transformer, graph, and tree models were trained on 1,849 actives and 34,503 inactives and used to prospectively score 6,747 drugs for activity against S. pneumoniae.
  • From 11 model-prioritized candidates tested in vitro, nine strongly reduced growth of S. pneumoniae R6 with IC50 ≤ 0.4 µg/mL.9/11 candidates active; IC50 ≤ 0.4 µg/mL
“we leveraged ensembles of transformer, graph, and tree models, each trained on a set of 1849 actives along with 34 503 inactives, to prospectively examine 6747 drugs.”
What this piece can’t prove

3 further details could not be confirmed from the summary.

2in vitroExperimentally validate AI-prioritized candidates with in vitro S. pneumoniae growth inhibition assays and estimate potency (IC50), identifying highly potent hits (e.g., thiostrepton, ceftiofur).in vitro growth inhibition (dose–response) assayExpand

In plain English

In vitro phenotypic validation of 11 AI-prioritized candidate antibiotics against Streptococcus pneumoniae R6 identified nine compounds that strongly reduced bacterial growth (IC50 ≤ 0.4 µg/mL). Two top hits, thiostrepton and ceftiofur, showed extremely low IC50s (0.0001 µg/mL and 0.0004 µg/mL, respectively). Thiostrepton retained high potency against multidrug-resistant S. pneumoniae strains.

Key findings

  • Screen of 11 AI-prioritized candidate antibiotics against S. pneumoniae R6 found nine compounds that strongly reduced in vitro growth (IC50 ≤ 0.4 µg/mL).IC50 ≤ 0.4 µg/mL (for nine validated compounds)
  • Thiostrepton and ceftiofur were the most potent hits, with extremely low IC50 values.Thiostrepton IC50 = 0.0001 µg/mL (60.1 pM); Ceftiofur IC50 = 0.0004 µg/mL (764 pM)
“Of 11 selected candidate antibiotics, nine were found to strongly reduce in vitro growth of S. pneumoniae R6, with IC50 values of ≤ 0.4 µg/mL.”
What this piece can’t prove
  • Abstract does not report experimental details such as assay conditions, number of replicates, strain panel composition, or statistical uncertainty for IC50 estimates.

2 further details could not be confirmed from the summary.

3in vitroAssess whether the top hit(s), especially thiostrepton, retain activity against multidrug-resistant S. pneumoniae strains, supporting potential stewardship-relevant use in non-invasive infection.in vitro susceptibility testing across MDR strain panelExpand

In plain English

The abstract reports that thiostrepton "remained highly potent even against multidrug-resistant strains" of Streptococcus pneumoniae in in vitro growth inhibition assays. The paper does not provide numeric potency values, strain identities, or assay parameters for the multidrug-resistant (MDR) panel in the abstract.

Key findings

  • Thiostrepton retained high in vitro potency against multidrug-resistant Streptococcus pneumoniae strains as reported in the abstract.
“Thiostrepton remained highly potent even against multidrug-resistant strains”
What this piece can’t prove
  • In vitro findings may not translate to clinical effectiveness or appropriate stewardship deployment without further preclinical and clinical data.

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.

Papers considered

The selected paper, plus nearby candidates.

PubMed, Europe PMC, Crossref · 15 candidate papers

And 9 more candidates considered.