AI drug design · Oncology imaging · Materials Today Bio
AI-designed PARP-1 PET tracer reaches first human scans in breast and ovarian cancer (Materials Today Bio)
Fudan University Shanghai Cancer Center used a generative-chemistry and docking pipeline to design the gallium-68 tracer [68Ga]Ga-DOTA-FZPF, then took it through mouse PET and an exploratory scan in 4 patients with breast or ovarian cancer. The authors say it is the first AI-designed radiotracer evaluated clinically.
Caveat: 4-patient exploratory study; imaging agent only, not a treatment; the authors call for larger studies.
Scienmag →
Longevity · Senescence · WIRED
Rubedo’s topical senescence-targeting cream lowered skin ’biological age’ by about 2 years in a small trial (WIRED, Oct 12)
In 11 people with atopic dermatitis, 4 weeks of the cream lowered skin age by about 2 years on average, measured by an independent Mitra Bio aging clock. The company also reported lower inflammation markers and about 10% more of one collagen type.
Caveat: Company data presented at the ARDD conference, not peer-reviewed; tiny, in eczema patients rather than healthy aging skin; aging-clock surrogate, not a clinical outcome.
WIRED →
Clinical AI · Thoracic oncology · JITC
MD Anderson’s CIPHER CT foundation model flags checkpoint-inhibitor pneumonitis risk before treatment (JITC)
Pretrained on about 590,000 CT slices from 2,500 lung cancer patients, the model reached an AUC of about 0.83 on routine pre-treatment CT in both the MD Anderson cohort (347) and an external Johns Hopkins cohort (116), beating clinical-factor and radiomics models.
Caveat: Retrospective; not clinically deployed; needs prospective validation.
MD Anderson →
JITC →
Clinical AI · Precision immunotherapy · Nature Medicine
I3LUNG: explainable AI on routine clinical and blood data beats PD-L1 for predicting immunotherapy benefit in advanced NSCLC (Nature Medicine)
In a 2,396-patient, 6-country real-world study, clinical and blood models reached AUC up to 0.77 in the test set and outperformed PD-L1, ECOG, NLR, LDH and LIPI. With the explainable tool, physicians’ sensitivity for predicting disease control rose from 0.72 to 0.87. Published in September; covered again Oct 11.
Caveat: External-validation AUC fell to 0.55–0.72; benefit of adding imaging and pathology is uncertain; prospective validation (NCT05537922) still underway.
Nature Medicine →
ClinicalTrials.gov →
Genes & disease · Multi-omics · Nature Communications
Multi-omics panel diagnoses pediatric mitochondrial disease and predicts who worsens (Nature Communications, Oct 9)
A 7-feature panel (6 metabolites, 1 transcript) reached AUC 0.96 in an independent validation cohort, better than conventional biomarkers. Proteomic clustering found two subtypes that diverged over time, and a 12-transcript signature predicted 12-month prognosis (AUC 0.85).
Caveat: Research biomarkers, not a cleared test; need broader validation before clinical use.
Nature Communications →
